From ffe817542e7f035c8fa70eef064b9a869d018cea Mon Sep 17 00:00:00 2001
From: rosalub <rosalie.dauchez.etu@univ-lille.fr>
Date: Wed, 3 Apr 2024 14:29:08 +0200
Subject: [PATCH] Adapt FederatedServer to use the new version of FLRunConfig
---
declearn/main/_server.py | 15 +-
quit | 7089 ++++++++++++++++++++++++++++++++++++++
2 files changed, 7097 insertions(+), 7 deletions(-)
create mode 100644 quit
diff --git a/declearn/main/_server.py b/declearn/main/_server.py
index 494bc3f7..7d921fed 100644
--- a/declearn/main/_server.py
+++ b/declearn/main/_server.py
@@ -205,9 +205,10 @@ class FederatedServer:
round_i = 0
while True:
round_i += 1
- await self.training_round(round_i, config.training)
- await self.evaluation_round(round_i, config.evaluate)
- if not self._keep_training(round_i, config.rounds, early_stop):
+ await self.training_round(round_i, config.get_next_train_config)
+ if config.get_next_valid_config is not None:
+ await self.evaluation_round(round_i, config.get_next_valid_config)
+ if not self._keep_training(round_i, early_stop, config):
break
# Interrupt training when time comes.
self.logger.info("Stopping training.")
@@ -387,10 +388,10 @@ class FederatedServer:
self.logger.info("Sending privacy requests to all clients.")
assert config.privacy is not None # else this method is not called
params = {
- "rounds": config.rounds,
- "batches": config.training.batch_cfg,
- "n_epoch": config.training.n_epoch,
- "n_steps": config.training.n_steps,
+ "rounds": config.n_rounds,
+ "batches": config.train_cfg.batch_cfg,
+ "n_epoch": config.train_cfg.n_epoch,
+ "n_steps": config.train_cfg.n_steps,
**dataclasses.asdict(config.privacy),
} # type: Dict[str, Any]
message = messaging.PrivacyRequest(**params)
diff --git a/quit b/quit
new file mode 100644
index 00000000..f3f031cc
--- /dev/null
+++ b/quit
@@ -0,0 +1,7089 @@
+[33mcommit 39bd3b932b3926093f8786491391b2b2c48653ef[m[33m ([m[1;36mHEAD -> [m[1;32mfedalgo[m[33m)[m
+Author: rosalub <rosalie.dauchez.etu@univ-lille.fr>
+Date: Mon Apr 1 13:07:27 2024 +0200
+
+ Incorporate a keep_training method in FLRunConfig and its child class, adapt it for FederatedServer
+
+[33mcommit 6f973ecfda2b80a943af9971bb5ef4138cd9369a[m[33m ([m[1;31morigin/fedalgo[m[33m)[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 29 16:16:13 2024 +0100
+
+ Start ammending new FLRunConfig API.
+
+[33mcommit 606833e385bee9338447f42316d8f65875d31bc9[m
+Author: rosalub <rosalie.dauchez.etu@univ-lille.fr>
+Date: Thu Mar 28 16:18:35 2024 +0100
+
+ Created concrete class for the basic federated learning approach which heritates from FLRunConfig
+
+[33mcommit aeb8e7a7bfc01479a95520470d31923d1e12723f[m
+Author: rosalub <rosalie.dauchez.etu@univ-lille.fr>
+Date: Wed Mar 27 16:25:35 2024 +0100
+
+ Abstraction of FLRunConfig
+
+[33mcommit 5881c22a437fb8c6f15751d058d1a7937d8d1d3b[m[33m ([m[1;32mdevelop[m[33m)[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 18 15:07:39 2024 +0100
+
+ Fix typos in release notes.
+
+[33mcommit c13246c293753c92f22de2bca30043f46d32fee4[m[33m ([m[1;33mtag: v2.4.0[m[33m)[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 18 15:02:02 2024 +0100
+
+ Release DecLearn v2.4.0.
+
+[33mcommit ff404f9f6ff1d85da3b3c19199a7987d27f72178[m
+Merge: 18a459f ffe6a30
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Mar 18 14:16:44 2024 +0100
+
+ Merge branch 'prerelease-2.4' into 'develop'
+
+ Finalize DecLearn v2.4.0.
+
+ See merge request magnet/declearn/declearn2!63
+
+[33mcommit ffe6a302a70c6dade9200e2c678a30c30756a6f8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 15 13:58:21 2024 +0100
+
+ Fix Torch 1.13 unit tests launch.
+
+[33mcommit aff0ff3185113a97b6603092b79ba9d30659541c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 15 11:29:16 2024 +0100
+
+ Add support for TensorFlow 2.16 / Keras 3.
+
+ - Apply some minor backward changes (and, when required, branches) to
+ support breaking changes in Keras 3 (the default for TensorFlow 2.16
+ and onwards), while preserving support for older versions.
+ - On the side, split a test about 'Model' serialization in two tests,
+ one for config serializability, the other for instantiation from a
+ non-serialized config dict.
+ - The current code has been tested (by running all unit tests) with
+ both TensorFlow 2.11 and 2.16, in a Python 3.10 environment.
+
+[33mcommit dce82854b3304a9a51a7332fbc24dfe9a27cefd0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 13 16:21:13 2024 +0100
+
+ Deprecate 'declearn.dataset.load_from_json'.
+
+[33mcommit 113f68df6245a7146677879ae061a7a46476228f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 13 16:02:38 2024 +0100
+
+ Update website documentation and enhance that of new Aggregate ABCs.
+
+[33mcommit 8318d23e4e211aff29f36b1d1bbf62d63577a0fc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Mar 12 16:07:19 2024 +0100
+
+ Add and deploy new utils for messages' type-checking.
+
+ - Add new 'declearn.communication.utils' submodule.
+ - Move existing utils there, but keep re-exporting them from the
+ parent level to avoid breaking existing code.
+ - Refactor some backend code used to verify exchanged messages'
+ type as part of federated learning processes into two new utils.
+
+[33mcommit 0b9edf38d8edaad0ada4351a05e5c031c1405272[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Mar 12 12:27:31 2024 +0100
+
+ Fix legacy 'Error' sending in edge case by 'GrpcServer'.
+
+[33mcommit 08b53632a288cf48a8d8553fc10570a0fbd524b0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 1 15:19:15 2024 +0100
+
+ Update release notes for '2.4.0'.
+
+[33mcommit d0b8eb2c7ec889db136ae097b7c757c72f6f5f2d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 1 15:08:39 2024 +0100
+
+ Fix redundant model weights sharing with clients.
+
+ - Until now, in the main FL process, clients would always receive
+ global model weights twice, due to their being shared both with
+ training and evaluation requests.
+ - This commit fixes that via the introduction of backend mechanisms
+ to keep track of which clients hold the latest model weights, and
+ sparingly attach them to requests based on this.
+ - As a side effect, `TrainRequest` and `EvaluationRequest` messages
+ now specify `weights` to be an `Optional[Vector]` field.
+
+[33mcommit 7704c46254eec0eaf5dcdf4ce1f8586cd7a893c2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 29 18:03:07 2024 +0100
+
+ Hack around PyLint not detecting 'tensorflow.keras' anymore.
+
+ This commit introduces imports of 'tf.keras' as 'tf_keras' so that
+ false-positive pylint errors can be silenced in one place per file
+ rather than on each and every call.
+ This remains tiresome, but might in fact be useful to switch to the
+ newer TensorFlow versions in the future.
+
+[33mcommit 196cef332de90fc8e0a41145e2d0539c9f425ac3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 29 17:30:18 2024 +0100
+
+ Update TensorFlow (optional) dependency version specification.
+
+ Set minimal version to 2.11.0 (rather than 2.5.0):
+ - 'TensorflowOptiModule' adapter is based on the "new" Optimizer API
+ introduced with TensorFlow/Keras v2.11, but the required dependency
+ version had not been updated until now.
+
+ Set maximal version to 2.15.X:
+ - TensorFlow 2.16 (the first release candidate of which was added to
+ PyPI three days ago) is switching to Keras 3 by default, which is
+ bound to introduce breaking changes (current unit tests fail with
+ 'tensorflow==2.16.0rc0').
+ - For now, the maximal version is thus set to 2.15.X. This may change
+ in the near future when some time and effort is put in understanding
+ changes and/or configuring dependencies to keep using Keras 2 rather
+ than Keras 3 in TensorFlow's backend.
+
+[33mcommit 921d17585deee665a82dbe853db336cace3bc2f9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 19 15:14:42 2024 +0100
+
+ Fix exceptions reporting in Heart UCI example.
+
+[33mcommit 7a8563a585af26dc2d4df3e9c456b7aa9887a4dc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 19 15:12:31 2024 +0100
+
+ Improve 'run_as_processes' backend to avoid specific kwargs loss.
+
+ - The previous implementation would result in dropping some kwargs
+ (namely, 'name' and 'queue') when running a function as process.
+ - The new backend implementation prevents this using a wrapped
+ instead of a partial function (still relying on 'functools').
+
+[33mcommit 7715ab915b04bfb468f2a49348cab255522ccfbf[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 19 11:49:39 2024 +0100
+
+ Write up release notes for 'v2.4.0'.
+
+[33mcommit e7fae4b5f9c39d7cdd4fc18dcb06e2ee80918e55[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 19 10:54:56 2024 +0100
+
+ Update online documentation to reflect API changes.
+
+ - Document changes in the registration and initialization processes.
+ - Document changes of the Optimizer AuxVar API.
+ - Document requirement for peers using the same DecLearn version.
+
+[33mcommit 542f5ad8dfaf990d244fc66f7c2a81e3d876fc7f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 19 10:14:10 2024 +0100
+
+ Fix docs' automated rendering.
+
+[33mcommit 6300a10e6c439874caf6373fd0f79924059cb66b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 16 17:54:02 2024 +0100
+
+ Remove deprecated features as planned since v2.2.
+
+[33mcommit e90280981de288587ee6c19506bc819d623e3e83[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 16 17:47:39 2024 +0100
+
+ Bump version to '2.4.0'.
+
+[33mcommit fc20c9ddf9765ef7f652fff654be4f6e37cd4b0c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 16 17:45:39 2024 +0100
+
+ Update 'black' version to '2.4.*'.
+
+[33mcommit 0d78e2c91fab00eabca798fbfa84e977d336ed76[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 16 17:37:03 2024 +0100
+
+ Quit using a scikit-learn deprecated attribute.
+
+[33mcommit 18a459fe653dfc2da507a09f6d5eeea1b576760a[m
+Merge: 7c65a28 f770f37
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Feb 16 17:52:31 2024 +0100
+
+ Merge branch 'revise-messaging' into 'develop'
+
+ Revise network communication backend and message-parsing logic.
+
+ See merge request magnet/declearn/declearn2!62
+
+[33mcommit f770f37037ba593dd4ccf121c3ea651e9eee6088[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 16 15:25:19 2024 +0100
+
+ Add 'heartbeat' parameter to 'NetworkServer' and speed up tests.
+
+ Set up a low heartbeat for localhost-running tests, resulting in
+ drastic runtime gains (notably due to time gains during the join
+ and initialization phase for functional tests).
+
+[33mcommit 07ed21db1f83cbf31d3b39ce4f291c8076696cc0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 16 15:01:39 2024 +0100
+
+ Add unit tests for 'declearn.communication.api.backend' code.
+
+[33mcommit 395436597f5a50b2e65c0d4e1aa5d45a00b501fc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 16 15:01:01 2024 +0100
+
+ Turn version-based rejection into a (version-based) constant flag.
+
+[33mcommit 7937bee773c9765a559568e0f661a0e46de5510a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 16 14:58:35 2024 +0100
+
+ Enable specifying message-related timeouts as floats.
+
+[33mcommit bbe43fba4e8c61a59057e8d518143b7f56453b9c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 15 17:52:46 2024 +0100
+
+ Update unit tests for 'communication' module.
+
+[33mcommit 1f554f141566574d9bacc8c03ed98a59a329f84c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 15 15:29:53 2024 +0100
+
+ Fix detection and rejection of legacy clients' messages.
+
+[33mcommit 3b8703f2cb18850d9ebcf9be37e38afe8fa0c845[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 15 12:09:06 2024 +0100
+
+ Adjust 'FederatedClient' and 'FederatedServer' to new messaging API.
+
+ - Update the way received messages are type-checked and have them
+ be deserialized after type-checking only.
+ - Move up metadata collection and use to a new optional and sparse
+ step during the initialization phase.
+ - Implement new message subtypes for metadata query and reply, and
+ for signalling that initializations steps went well.
+
+[33mcommit 125e2708881f527dd6a3a2301687f29f5aa7421f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 15 12:03:58 2024 +0100
+
+ Integrate new Message API with communication endpoints.
+
+ - Revert some changes from a previous commit, that may be envisioned
+ again when revising APIs in-depth for DecLearn 3.0.
+ - Stitch back the 'communication' and 'messaging' submodules together,
+ having 'NetworkClient' and 'NetworkServer' send 'Message' instances
+ (that are merely serialized to string prior to being exchanged) and
+ parse received strings into 'SerializedMessage' instances.
+ - This change is a compromise between keeping things as before (which
+ would not benefit from the introduced delayed-parsing capability of
+ 'SerializedMessage'), and changing them so deeply that current code
+ making use of communication endpoint would require heavy revisions.
+
+[33mcommit 4215381049c3d61de518a5c8f77099ec7839b2be[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Feb 13 17:23:55 2024 +0100
+
+ Revise 'Message' API, make it extendable and move it around.
+
+ * Introduce new 'declearn.messaging' submodule, that is meant to
+ replace the now-deprecated 'declearn.communication.messaging'.
+ * Make 'Message' extendable via subclassing and type-registration
+ (which is automatically performed by default).
+ * Introduce 'SerializedMessage' as a wrapper for received messages,
+ that parses the exact message subtype (enabling logic tests and
+ message filtering) but delays actual content deserialization and
+ Message object recovery (enabling to prevent undue resources use
+ for unwanted messages that end up being discarded).
+ * Deprecate some old 'Message' classes, that are still exported by
+ the deprecated submodule but not by the new one.
+
+[33mcommit 7b6233170727d89d10ff2a3b426b7085cdc6624a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Feb 13 15:15:20 2024 +0100
+
+ Revise network communication backend and message-parsing logic.
+
+ - Introduce a new 'declearn.communication.api.backend' submodule:
+ - Introduce a new 'ActionMessage' minimal backend API, that defines
+ hard-coded, lightweight and easily-parsed data structures designed
+ to convey information and content across network communications
+ agnostic to the content's nature.
+ - Move 'declearn.communication.messaging.flags' submodule to
+ 'declearn.communication.api.backend.flags'.
+ - Move 'MessagesHandler' semi-private class to the 'backend'
+ submodule, where it is now properly exposed.
+
+ - Revise 'MessagesHandler':
+ - Have it use the new 'ActionMessage's to wrap up messages, and
+ delay their content's parsing to other times and parts of the
+ code.
+ - Quit sending information together with registration requests.
+ In the future, we may want to introduce some callback system to
+ validate or complete a registration request's processing.
+ - Turn 'heartbeat' into an instantiation parameter.
+
+ - Revise 'NetworkClient' and 'NetworkServer':
+ - Quit passing of 'data_info' with registration requests.
+ - Have message-sending and message-receiving methods expect str
+ inputs and return str outputs: the actual parsing into objects
+ is therefore delayed and left to be defined elsewhere.
+ - Rename 'NetworkClient.check_message' into 'recv_message' (keep
+ the former as an alias, with a DeprecationWarning).
+ - Improve the use of (optional) timeouts when sending or expecting
+ messages and overall exceptions handling:
+ - 'NetworkClient.recv_message' may either raise a TimeoutError
+ (in case of timeout) or RuntimeError (in case of rejection).
+ - 'NeworkServer.send_messages' and 'broadcast_message' quietly
+ stops waiting for clients to collect messages after the (opt.)
+ timeout delay has passed. Messages may still be collected.
+ - 'NetworkServer.wait_for_messages' no longer accepts a timeout.
+ - 'NetworkServer.wait_for_messages_with_timeout' implements the
+ possibility to setup a timeout. It returns both received client
+ replies and a list of clients that failed to answer.
+
+[33mcommit 7c65a28361af1e2c31fcf1d2d32fac471c6207da[m
+Merge: d240079 03efb10
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Feb 9 13:04:01 2024 +0000
+
+ Merge branch 'revise-aggregator-api' into 'develop'
+
+ Introduce and deploy `Aggregate` API
+
+ See merge request magnet/declearn/declearn2!61
+
+[33mcommit 03efb109e6035c786d682a4d7fca6496ac8f1514[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 9 12:06:23 2024 +0100
+
+ Add 'declearn.version' namespace to avoid circular imports of '__version__'.
+
+[33mcommit 47cc28436188e1f62cd6c5279c13636b68783d8f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 25 15:14:30 2024 +0100
+
+ Add unit tests for all 'Aggregate.prepare_for_secagg' methods.
+
+[33mcommit 71d1e62407fd4544e501dc825f61eedf8ebf7d12[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 25 11:00:02 2024 +0100
+
+ Expand 'Aggregate' API to specify how to use SecAgg on top of it.
+
+[33mcommit fd01dfd784c69f41e177a29a685156311563ddbd[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 25 10:53:06 2024 +0100
+
+ Make 'VectorSpec' JSON-serializable.
+
+[33mcommit b12c8d8a1cfe06ebb73a5031bf216978c669ddc0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Oct 19 17:50:44 2023 +0200
+
+ Expand 'Vector' API to enable (un)flattening to/from list of floats.
+
+ - Add `VectorSpec` dataclass to store metadata about a Vector.
+ - Add `Vector.get_vector_specs` to access the metadata of an instance.
+ - Add `Vector.flatten` and `Vector.unflatten` to convert to and from
+ a list of floats and a VectorSpec instance.
+ - Add `Vector.build_from_specs` static method as a generic entrypoint
+ to unflatten any Vector instance from its list of float values and
+ VectorSpec instance, provided it was properly type-registered.
+ - Implement these new methods for each and every `Vector` subclass
+ currently shipped with declearn.
+ - Add unit tests for these features as part of the standard test suite
+ applied to each and every type-registered `Vector` subclass.
+ - All of these new features are designed to enable SecAgg of Vector-
+ wrapped values, which requires operating on (very) big integers
+ that would result in overflows in most backend data frameworks,
+ starting with numpy.
+
+[33mcommit d40233861a1eb3ece19ca32e0c9309caedc9ab65[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 23 16:10:26 2024 +0100
+
+ Update 'BinaryRocAUC' to better handle boundedness of thresholds.
+
+[33mcommit 389d58ba6b088f660b4d6d998c6adc4763d618ec[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jan 22 13:22:55 2024 +0100
+
+ Update unit tests for 'Metrics' and revise/fix associated code.
+
+[33mcommit 9141b0403e2d49e3bccd2d0e8d5fe96966ff68d6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 19 16:08:07 2024 +0100
+
+ Revise 'Metric' API introducing 'MetricState' aggregates.
+
+[33mcommit bc74f367609a680232b366caccc15286087277ac[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 19 17:37:01 2024 +0100
+
+ Update 'declearn.optimizer' unit tests.
+
+[33mcommit 3ee402cb960a9133f63255820718cf1947dc2a97[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 23 12:06:30 2024 +0100
+
+ Implement declearn version verification when registering clients.
+
+ * Clients now send their local declearn version string as part of
+ their registration request to the server.
+ * Server now rejects requests that have unmatching 'major.minor'
+ version string.
+ * This may be revised in the future to implement an asymmetric
+ requirement where clients may have more recent minor versions
+ than the server. For now, this will however lessen the burden
+ of having to deal with complicated retro-compatibility issues.
+
+[33mcommit 4262f78840710d87fe5b39d85461a363f6d2fd38[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 19 12:11:07 2024 +0100
+
+ Abstract new 'Aggregate' ABC from 'AuxVar' and 'ModelUpdates'
+
+[33mcommit 7c1baf4f2c36b94ebab2ce7342d2b935b349c7d4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 19 11:29:53 2024 +0100
+
+ Preserve 'AuxVar' and 'ModelUpdates' types across network messaging.
+
+ * Automate the extension of JSON support to 'AuxVar' and 'ModelUpdates'
+ (sub)classes.
+ * Revise 'TrainReply' and 'TrainRequest' to pass objects directly rather
+ than their dict representation.
+ * Overload these messages' 'to_kwargs' methods to avoid recursive dict
+ conversion of dataclasses, hence ensuring that types are preserved
+ and rebuilt upon messages' deserialization.
+
+[33mcommit 1c977320d574945648f2fe12fd8a60d33907c50c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 19 10:41:48 2024 +0100
+
+ Refactor 'Message.to_string', adding 'Message.to_kwargs'.
+
+ - Refactor 'Message.to_string' into a pair of methods:
+ - 'to_kwargs' to produce a JSON-serializable dict of kwargs
+ - 'to_string' that calls the former, dumps to JSON and adds
+ the message's typekey.
+ - Note: in the future we may change things to have the typekey
+ parsable without deserializing the rest of the message, and
+ then add some intermediate structure where deserialization
+ is yet to occur, so that message validity can be checked prior
+ to deserialization of its content.
+
+[33mcommit 80865e116dbe297740b9f5f08cb9fdf58c04206b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 18 10:49:53 2024 +0100
+
+ Deploy new 'AuxVar' to the federated learning process.
+
+ * Have 'FederatedServer' and 'FederatedClient' (via 'TrainingManager')
+ convert emitted auxiliary variables to dict representations.
+ * Also have them reconstruct 'AuxVar' structures upon reception,
+ using the newly-introduced 'Optimizer.unpack_aux_var' method.
+ * Rewrite the aggregation of clients' auxiliary variables, taking
+ advantage of the new 'AuxVar' structure, and pass the resulting
+ one-object-per-module to the server's 'Optimizer'.
+
+ Notes:
+ * The all 'unpack_aux_var' may be dropped in favor of preserving
+ 'AuxVar' types across serialization (and thus across network),
+ using the usual type-registration and JSON-extending systems.
+
+[33mcommit bd97e31427f32949571b65387111af856f818362[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 17 16:55:03 2024 +0100
+
+ Revise OptiModule Auxiliary Variables, and Scaffold implementation.
+
+ - Replace dictionaries with dataclass structures that may implement
+ aggregation rules (with a default of sum-aggregation).
+ - Drop support for client-wise values emission from the server, and
+ move aggregation outside of the 'Optimizer' class.
+ - Rewrite Scaffold to make it more efficient, and SecAgg-compatible
+ (with clients sending Vector information that are to be summed).
+
+ Notes:
+ - Actual aggregation of new-type auxiliary variables is yet to be
+ implemented. The same design questions arise as with 'Aggregator',
+ and the final answers should be the same for tidyness purposes.
+ - Tests are yet to be revised and run for debugging purposes.
+ - 'AuxVar' code is largely redundant with 'ModelUpdates' and should
+ be abstracted into some common ancestor for maintainability.
+ - At first, some retro-compatibility was considered: newer Scaffold
+ clients would be able to communicate with older servers. However,
+ the opposite seems not to be (easily) achievable. In the end, this
+ was dropped, as it would cause maintenance costs and hurt code
+ readability for no practical benefit given the current user-base.
+
+[33mcommit cc8a19ef921a1eeaed083685bb0ec667a96d9c77[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 17 11:09:25 2024 +0100
+
+ Update unit tests for 'Aggregator' to cover new API methods.
+
+[33mcommit 3428ab41b2a0b840185383431233b265117a8e5c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 17 10:31:45 2024 +0100
+
+ Deploy new 'Aggregator' API to the main federated learning process.
+
+ * Add 'Aggregator' to objects shared with clients in 'InitRequest'.
+ * Add 'Aggregator' to the 'TrainingManager' and make use of it to
+ prepare and package shared model updates.
+ * Use 'Aggregator.updates_cls' to recover 'ModelUpdates' objects
+ from client-emitted dict representations, sum-aggregate them
+ and finalize them using 'Aggregator.finalize_updates' on the
+ server side.
+ * Fix/update 'TrainingManager' unit tests based on changes.
+
+ Notes:
+ * 'TrainingManager' is due for a revision, that will probably
+ result in splitting the training loop away from results'
+ packaging for sharing.
+ * 'ModelUpdates' may be updated with type-registration tools,
+ so that its type is recovered upon deserialization rather
+ than through an explicit call to 'Aggregator.updates_cls'.
+
+[33mcommit 137f02308634021a62cbded3f4b95b7211397756[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 16 16:28:25 2024 +0100
+
+ Revise 'Aggregator' API, replacing 'aggregate' with new methods.
+
+ * Revise 'Aggregator' API entirely:
+ - Define 'ModelUpdates', a dataclass holding (partial) aggregates
+ for their sharing and aggregation across peers.
+ - Add 'Aggregator.prepare_for_sharing' to the API, that is bound
+ to prepare and package local updates for their sum-aggregation.
+ - Add 'Aggregator.finalize_aggregates' to the API, that is bound
+ to compute a 'Vector' of global model updates from a single
+ 'ModelUpdates' instance (holding pre-aggregated values).
+ - Deprecate 'Aggregator.aggregate' API method, but make it work
+ in legacy mode using new methods as core backend.
+ * Rewrite 'AveragingAggregator' and 'GradientsMaskAveraging'.
+ - 'AveragingAggregator' no longer supports using client weights,
+ save when using the deprecated (legacy) 'aggregate' method.
+ - 'GradientsMaskAveraging' no longer inherits 'AveragingAggregator',
+ but uses it by composition instead, notably to avoid type-hinting
+ issues.
+ - 'GradientsMaskAveraging' is designed to compute updates' sign with
+ parsimony, to avoid information duplication across network in the
+ non-secagg federated setting.
+
+ * Notes for the future:
+ - 'ModelUpdates' could be made to inherit from a much more generic ABC,
+ that would enable writing aggregation rules for many data containers.
+ - It is yet unsure how/whether the 'aggregates_cls' class attribute is
+ going to be used; it may be removed during development.
+ - SecAgg is yet to be plugged into this code. The current rationale is
+ to write dedicated code at the level of 'ModelUpdates' to encrypt or
+ decrypt content - and it might require some dedicated wrapper type.
+
+[33mcommit d24007973b7df864f4b2bf3d3a1ea6fe9cd237fa[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 30 15:59:51 2024 +0100
+
+ Marginally improve 'run_tests.sh' commandline outputs.
+
+[33mcommit 7649b725519742d344b5c60ceabe06e07b068f91[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jan 29 16:14:45 2024 +0100
+
+ Sort internal imports under 'declearn.utils'.
+
+[33mcommit c450f6af5cf059c2376b68169d07b76766121e1d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 26 11:04:22 2024 +0100
+
+ Add a shared '.gitignore' file.
+
+[33mcommit 428cf933ab9310c02820faee3523095b3c42f1fe[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 11 16:59:58 2024 +0100
+
+ Fix 'BinaryAccuracyPrecisionRecall' computations without true positives.
+
+[33mcommit 6497f761e111959ed3aa4ec86c089d8723d6fd8e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Dec 14 11:56:11 2023 +0100
+
+ Fix some 'pragma: no cover' comments for typos.
+
+[33mcommit daef0c0a4b2ff10aed2ae3e543d8eee67a485704[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Dec 14 11:28:04 2023 +0100
+
+ Remove pylint-false-positive-silencing comments regarding torch.
+
+ False-positives on many `torch.<something>` symbols appear to have
+ been fixed. Whether this comes from a Pylint and/or Torch update
+ is unclear, but as developers are expected to work with up-to-date
+ installations (and so does the CI), we can safely remove silencing
+ comments.
+
+[33mcommit 73a8f80afcc9beeef39ba12c3303cc6a09bd8377[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Dec 14 11:26:39 2023 +0100
+
+ Fix 'TorchVector.(min|max)imum' failing on GPU with float input.
+
+[33mcommit 33dca7cd0b00ed8befd108f24169274938a0b64c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Dec 4 13:47:13 2023 +0100
+
+ Change pandas version specifier to re-enable v2.x.
+
+[33mcommit 5c1fe0fd056a96fe2d3b63ba21a6e7b7178c77e7[m
+Merge: 6226ba3 1931b16
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Dec 4 12:44:18 2023 +0000
+
+ Merge branch 'fix-ci-gpu' into 'develop'
+
+ Enable GPU use on our CI runner and garden around 'pyproject.toml'.
+
+ See merge request magnet/declearn/declearn2!60
+
+[33mcommit 1931b166a705255d4fd21748340fc510d32f1890[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 23 15:06:35 2023 +0100
+
+ Add support for recent versions of websockets.
+
+ - The 'websockets' dependency has been known to introduce breaking
+ changes with its major version, hence our sticking with v10 since
+ DecLearn was first implemented.
+ - However, the two latest major version releases are documented and
+ have been tested not to break compatibility with our code; hence
+ we now explicitly support them as compatible dependencies.
+
+[33mcommit df0829b3bcd440135309e66171ec5425de141ce8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 23 15:00:50 2023 +0100
+
+ Enable using the more recent versions of dm-haiku.
+
+ - Haiku is still at the early stages of its developments, hence
+ we are being somewhat conservative in its version specifier.
+ - However, the latest versions prove not to break compatibility
+ with Python >=3.9, hence we now explicitly support them.
+ - For Python 3.8, the latest is no longer compatible, due to the
+ use of some newer type-hint syntax. We hence use version-based
+ specifiers.
+
+[33mcommit 07ea0a24f371f22aeac8abbcae870eb90b0bc9ea[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 23 15:00:16 2023 +0100
+
+ Improve dependencies version specifiers, using '~=' in most places.
+
+[33mcommit 549b77667c5d1c0169c32000f460a138d45012a7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 22 10:21:19 2023 +0100
+
+ Update 'tox.ini' to improve GPU usage.
+
+ - Add environment variables to prevent over-allocation of GPU VRAM
+ by TensorFlow and/or Jax.
+ - Add 'py{3x}-ci' test environments that are tweaked specifically
+ for our CI runner, adding instructions that offer as much support
+ for GPU as possible given our runner host's configuration.
+ - At the moment, an update of the host to the latest CUDA version
+ enables native support for Torch, and TensorFlow with Python>=3.9
+ (but not 3.8), plus Jax with some extra installation command.
+
+[33mcommit 2743d0d24409e112e4cc481704c1ceade66a2847[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 23 14:43:02 2023 +0100
+
+ Stop forcing '--full-trace' mode of pytest.
+
+[33mcommit d50bc747e403ee62e189cd5f52531b8059f19d3e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 23 14:15:14 2023 +0100
+
+ Fix some minor linter warnings on unit tests' code.
+
+[33mcommit 4a21f58712399e2063c00341a8d5267f31a1faf6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 23 11:49:32 2023 +0100
+
+ Fix the way GPU-use is disabled for some unit tests.
+
+[33mcommit e84c69c3fea72f8eefd3bdafcb380ecdadf89a1e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 23 11:42:54 2023 +0100
+
+ Fix tests' automated type-checking.
+
+[33mcommit 48711aa9eb119fa64b051832ae86aeb6518ed0d5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 23 11:42:24 2023 +0100
+
+ Complete docs and type-hints of 'test_multiprocessing'.
+
+[33mcommit 7adbbebfd6881ba7b29299b2c029acad9da8ce86[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 23 10:59:12 2023 +0100
+
+ Fix 'JaxNumpyVector' equality operator on distinct-device inputs.
+
+[33mcommit afbae61520ae2e2e0a1e8d0ea8ba697338cb095c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 22 12:33:29 2023 +0100
+
+ Ensure GPU is unused in toy-regression functional tests.
+
+[33mcommit eabba124d56a1f78b6eb605b8c5cd9d1059d9f3c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 22 11:39:50 2023 +0100
+
+ Fix TensorFlow Adam optimizer unit test.
+
+[33mcommit 6226ba34f82422754acc57423b8bf37ea89952dc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 7 11:42:52 2023 +0200
+
+ Add a guide about the Optimizer API to the docs.
+
+[33mcommit ef7f4bcbbe955d179d7c89093c374f99054e8de9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 7 10:52:24 2023 +0200
+
+ Improve website code blocks' rendering.
+
+[33mcommit 6d935012c5302188623b34d91f0f0a965032fa11[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 29 14:06:12 2023 +0100
+
+ Improve a docstring of ScaffoldClientModule.
+
+[33mcommit f2c14debfe028624e5fecb4c574602b1e0816a0e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 22 10:20:36 2023 +0100
+
+ Fix device-placement of jax-numpy stub data by 'GradientsTestCase'.
+
+[33mcommit c148e388dd99d58499b11386d750e23a1158204a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 22 10:18:12 2023 +0100
+
+ Improve torch dependencies management in the tests-running script.
+
+[33mcommit d99c74908a86ecc04f4a3efc1857f7a530e56c9b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 22 10:16:10 2023 +0100
+
+ Update 'opacus' dependency required version to '~=1.4'.
+
+ Former versions of opacus require functorch, which is eager to
+ cause some conflicts with recent torch versions.
+
+[33mcommit 83ada7a5b1f8e92e46b00bedfb5f7d31db3453e1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 22 10:08:24 2023 +0100
+
+ Correct syntax and comment 'jax.config' calls in tests.
+
+[33mcommit f48cec58de25ae091d481fb82a9702327c0e3586[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 22 10:05:42 2023 +0100
+
+ Minor mypy-based type-check correction.
+
+[33mcommit 583e2fc785520c202b25a53c65aaa02e5d2b53ad[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Oct 30 16:16:11 2023 +0100
+
+ Fix coherence of torch dependencies specification.
+
+[33mcommit 6c36112f4a8fda32285d623f27251ce6583e7ec1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Oct 10 16:30:35 2023 +0200
+
+ Upgrade Pylint requirement to version 3.
+
+[33mcommit cf016b5f28f761536dc87ce3cc9bf88f662caa16[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Oct 10 16:02:17 2023 +0200
+
+ Improve 'mnist_quickrun' example documentation and add torch variant.
+
+[33mcommit 7166e0a6194f41c9ed1cfa27d78b3c47aef82021[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Oct 10 16:01:24 2023 +0200
+
+ Revise 'declearn-quickrun' backend to use asyncio.
+
+[33mcommit 9ca2167727ae1a44d55fd867460ffccbb7df483d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Oct 5 11:48:50 2023 +0200
+
+ Add release notes for v2.3.2.
+
+[33mcommit cfbf09d9a0653d1731a5035517471bfb13373a23[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Oct 5 11:24:10 2023 +0200
+
+ Update 'torch' extra dependency specifiers.
+
+ - Drop support for Torch 1.10-1.12, which complexify rules for functorch.
+ - Make "torch", "torch1" and "torch2" specifiers deal merely with Torch,
+ leaving Opacus to the "dp" specifier, as should be.
+ - Have functorch be installed together with torch, removing some burden
+ from our specifier, and enabling support for the new Torch 2.1.
+
+[33mcommit edd2d777ff8e68f74afdab0cdc7ab480fb9ea258[m
+Merge: ad09f31 c379c5f
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Sep 26 13:53:04 2023 +0000
+
+ Merge branch 'display-progress-bars' into 'develop'
+
+ Add 'verbose' argument to 'FederatedClient' and 'TrainingManager'
+
+ See merge request magnet/declearn/declearn2!59
+
+[33mcommit ad09f3116a3accb2258d00f62287c17fffce705b[m
+Merge: a7e1b27 4067160
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Sep 26 13:36:08 2023 +0000
+
+ Merge branch 'record-training-losses' into 'develop'
+
+ Enable recording training loss values
+
+ See merge request magnet/declearn/declearn2!58
+
+[33mcommit 40671604c3f219f622f513fcb4edd776391443b8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 21 17:33:27 2023 +0200
+
+ Add (optional) training losses export to 'FederatedClient'.
+
+[33mcommit 753b43f693b583bc97f63547b1d824a18052c66a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 21 17:32:54 2023 +0200
+
+ Minor backend changes to 'Checkpointer'.
+
+[33mcommit 23726d573eeb503f1881f0eafccea627684ccd4a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 21 16:49:33 2023 +0200
+
+ Add 'TrainingManager.(train|evaluate)_under_constraints' to the API.
+
+ - Until now, the public methods for training and evaluation using
+ the `declearn.main.utils.TrainingManager` class (and its DP-SGD
+ counterpart) required message-wrapped inputs and emitted similar
+ outputs.
+ - With this commit, the previously-private `train_under_constraints`
+ and `evaluate_under_constraints` routines are made public and thus
+ part of the declearn API, with some minor refactoring to make them
+ more user-friendly.
+ - The rationale of this change is to enable using `TrainingManager`
+ outside of our `FederatedClient`/`FederatedServer` orchestration,
+ notably when simulating FL training or testing client-side code.
+ It may also be helpful to end-users that would like to build on
+ declearn but implement their own orchestration tools or algorithm
+ loops.
+ - In the future (declearn >=3.0), we may go one step further and
+ take all the messaging-related instructions out of the current
+ class. The class may also be moved to a different namespace, e.g.
+ a new 'declearn.train' module; but this is entirely out of scope
+ for now.
+
+[33mcommit c379c5fc187d2eaa5f3347e71a6a13dab01b77bb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 21 12:01:11 2023 +0200
+
+ Add 'verbose' argument to 'FederatedClient'.
+
+[33mcommit 03151d4938f3cdad974e2f00cd3a335b1f4f03d4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 21 11:59:55 2023 +0200
+
+ Add optional progress bars to 'TrainingManager'.
+
+[33mcommit 9a3cd4c78ec9004a254ee5c0de4b82f02936f745[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 21 11:51:50 2023 +0200
+
+ Add unit tests for 'run_as_processes'.
+
+[33mcommit 7ff01bc20465d534665d99749e9f983196141820[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 21 11:51:35 2023 +0200
+
+ Modularize 'run_as_processes' input specs.
+
+[33mcommit e820470ce966f9f2c80d8505ccde297c07603a0a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 20 14:46:51 2023 +0200
+
+ Implement training losses' recording in 'Model'.
+
+[33mcommit a7e1b273843919ed50636286e40bb70079ee7e18[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 6 13:50:28 2023 +0200
+
+ Update website index of release notes.
+
+[33mcommit 34ee4bfbd757f297086df6ece3e1bcfdc30e31d8[m[33m ([m[1;33mtag: v2.3.1[m[33m)[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 6 11:26:29 2023 +0200
+
+ Update 'tox.ini' as openssl is no longer called directly.
+
+[33mcommit 905dcc022d1eb442254766679e45497121ea1442[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 6 11:06:33 2023 +0200
+
+ Add release notes for v2.3.1 and v2.2.2.
+
+[33mcommit 911c67a9b89c05924800eb49566867d13b16246f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 6 11:05:47 2023 +0200
+
+ Bump version to '2.3.1'.
+
+[33mcommit 4de359e9734cc4cc372a2351d8e0ff2d29a0fa3d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 6 10:49:35 2023 +0200
+
+ Drop 'functorch.compile' use.
+
+ - The use of 'functorch.compile' over a function that takes variable-size
+ batch inputs proves impossible, as the tracing on first call creates a
+ computation graph with fixed dimensions.
+ - As a result, the tentative compilation of per-sample clipped gradients
+ computation prevents the proper use of DP-SGD with the functorch backend.
+ - An alternative attempt was to compile the sample-wise function and vmap
+ it afterwards, but this is currently unsupported (and unlikely to be as
+ part of functorch, as development efforts have moved to 'torch.func').
+ - This commit therefore drops the use of 'functorch.compile'.
+
+[33mcommit c65d0b4ca18039ec873f315a08d74afa7371acb6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Sep 1 11:57:00 2023 +0200
+
+ Improve release notes formatting.
+
+[33mcommit 783869c2e96794124e97d1d485b8488ac94cd024[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Sep 1 11:26:18 2023 +0200
+
+ Fix self-signed CA generation.
+
+[33mcommit eed7b8868f69694e615f3e15dc18e94378e9b3f7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 31 17:39:17 2023 +0200
+
+ Add a 'duration' parameter to 'generate_ssl_certificates'.
+
+[33mcommit 8f8199db154bac9939772ce07fc345ec23a4e119[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 9 16:34:05 2023 +0200
+
+ Rewrite 'generate_ssl_certificates' backend using 'cryptography'.
+
+[33mcommit 7cb704dc89c07ad726223d685a3ad763bb3299fc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 31 15:37:59 2023 +0200
+
+ Patch website docs.
+
+[33mcommit cba2534d450e0e29ad6ea97b426a63b0298334f4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 31 15:30:51 2023 +0200
+
+ Fix some internal documentation links.
+
+[33mcommit ad56d1e175e8e0a17a20ebdfdf66b9cf8f5b3e95[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 31 15:30:27 2023 +0200
+
+ Fix 'L2GlobalClipping' not being properly exported.
+
+[33mcommit 2a85461eade1bf7a87e6e68d998ad86b580c0a05[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 31 15:29:35 2023 +0200
+
+ Add release notes to the website docs.
+
+[33mcommit 8a0c4c8542612a0b44649a86b3882ecffeb79cb1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 31 15:29:08 2023 +0200
+
+ Document GitLab/GitHub repositories use.
+
+[33mcommit 03e7b1f8fbef914177bdea505b4a13a17ba662e7[m[33m ([m[1;33mtag: v2.3.0[m[33m)[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 31 09:28:50 2023 +0200
+
+ Bump version to 2.3.0 and update authors info.
+
+[33mcommit ad4814320883006f09615529eeb7a0f94f811c6b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 31 09:29:45 2023 +0200
+
+ Add a missing file license header.
+
+[33mcommit 094ab9eb2e7205b141f168e79c006eb2b4eb6a14[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 31 09:26:46 2023 +0200
+
+ Correct the docs of 'EvaluateConfig'.
+
+[33mcommit 899be96ea22f270c891b18d7bdcd8453122ece7d[m
+Merge: 3d39a2d ec75ca3
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Aug 30 14:24:38 2023 +0000
+
+ Merge branch 'gardening' into 'develop'
+
+ Improve tests coverage and fix test-digged bugs
+
+ See merge request magnet/declearn/declearn2!57
+
+[33mcommit ec75ca3239617a4f8993495ca64cd87cbacbbd53[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 16:10:52 2023 +0200
+
+ Rename same-name test files.
+
+[33mcommit 878057dae292e146e998eb927b2709a198c56089[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 15:46:01 2023 +0200
+
+ Update unit tests to use optimizer plug-in listing utils.
+
+[33mcommit b5121d0fc9b641a85cac63714ded88d86a1ae8cb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 15:43:25 2023 +0200
+
+ Add unit tests for 'aggregate_clients_data_info'.
+
+[33mcommit 6c9b983d49b6b5ea76a4cf53b6ce6df7671765fd[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 15:02:13 2023 +0200
+
+ Fix EarlyStopping for exact repeated inputs.
+
+[33mcommit 480a5995f21651c7e4b1f7bc5cd00696c7160e82[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 15:01:11 2023 +0200
+
+ Add unit tests for 'EarlyStopping'.
+
+[33mcommit 730db92c34a57d19fe9c1fa3ba73142c120a45fc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jun 8 11:35:44 2023 +0200
+
+ Add tests for 'declearn.quickrun.parse_data_folder'.
+
+[33mcommit 89c807bb916107f6ee0f113edf698f74b52c3cec[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jun 8 10:53:54 2023 +0200
+
+ Add unit tests for some quickrun backend utils.
+
+[33mcommit 474f544771669b045660e3ef12192905a789dd65[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 12:21:50 2023 +0200
+
+ Expand 'InMemoryDataset' unit tests.
+
+[33mcommit d8d078ee8b8076ed4e175bebf9485681894eeedf[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 12:21:46 2023 +0200
+
+ Adjust 'pragma: no cover' for 'InMemoryDataset'.
+
+[33mcommit 0c13373b9167d62bddd8672a3cc837da98b884b8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 12:20:55 2023 +0200
+
+ Deprecate 'InMemoryDataset' JSON-dump (private) type-key use.
+
+[33mcommit 87eada196f528d93620b2b0dcd8aebf4e611e752[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 12:20:24 2023 +0200
+
+ Fix 'InMemoryDataset.data_type'.
+
+[33mcommit 4d0226f3636f536a089e1ae6c2a50d19fc51bffa[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 12:19:37 2023 +0200
+
+ Fix 'InMemoryDataset' single-column target loading from csv.
+
+[33mcommit beae850833f658adeb2269d7e0615942e624e3a9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 30 10:29:51 2023 +0200
+
+ Add unit tests for high-level data info utils.
+
+[33mcommit 2a9a53134e91ead0dad7e097598b59c275437f2f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 25 16:23:42 2023 +0200
+
+ Rewrite toy regression integration tests.
+
+[33mcommit 811699cac6565e912c6abbb12935fd70c80fd082[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 25 10:27:03 2023 +0200
+
+ Fix expanded-dim inputs handling in MAE, MSE and R2 metrics.
+
+ Until now, when provided with y_true:<shape>, y_pred:(<shape>, 1)
+ inputs, the MAE, MSE and R2 metrics would be entirely wrong, due
+ to the way numpy casts operations between such inputs. This patch
+ adds some shape-verification and squeezing operations that fix the
+ computations. Unit tests were added to cover this case.
+
+[33mcommit cfb294ea8fd59de112fec0e7688d5120baa453af[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 24 14:21:44 2023 +0200
+
+ Update Jax / Haiku dependency specifications.
+
+[33mcommit d10a42e2de75a854882f941aa5d4d840e098eb83[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 24 13:54:23 2023 +0200
+
+ Deploy 'TensorflowDataset' in toy-regression integration test.
+
+[33mcommit 0a242469fa58273743a357f4d4107373dceaef2e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 24 11:27:56 2023 +0200
+
+ Fix 'DataTypeField.is_valid' and update deprecated fields' doc.
+
+[33mcommit 044466096d7d7b5ebd8dff2153d992c1f4a73e20[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 24 11:28:16 2023 +0200
+
+ Add unit tests for 'DataInfoField' subclasses.
+
+[33mcommit 44dbcd1c6def023fcb9ae893b41f940b1b676434[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 24 10:47:58 2023 +0200
+
+ Clean up some Websockets backend code.
+
+[33mcommit 91adc4077ca2bfc54429c5c4c1ee6ea053e5b579[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 24 10:47:26 2023 +0200
+
+ Add unit tests for large (chunked) messages exchange.
+
+[33mcommit 48d18343bf39746d2687bb8c017fe5ce20a28d06[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 9 17:04:18 2023 +0200
+
+ Add some 'pragma: no cover' for deprecated functions and edge-case exception handling.
+
+[33mcommit 6a18057ca219eb565c2f2e4d72396da24282ce4e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 9 17:03:52 2023 +0200
+
+ Correct v2.3-introduced deprecations' planned removal.
+
+[33mcommit 3d39a2d051e58eb7b482f301eeb5184635664700[m
+Merge: 62cc5a6 2795507
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Aug 9 12:36:38 2023 +0000
+
+ Merge branch 'global-clipping' into 'develop'
+
+ Implement 'L2GlobalClipping' OptiModule.
+
+ See merge request magnet/declearn/declearn2!56
+
+[33mcommit 2795507c8b9ad5a13ef84d4af876ce3096f5602f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 3 17:49:02 2023 +0200
+
+ Implement 'L2GlobalClipping' OptiModule.
+
+[33mcommit 62cc5a69d18169a74affbaf1b61aa175e900918f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 16:06:31 2023 +0200
+
+ Add missing seed to Torch and TensorFlow Dataset tests.
+
+[33mcommit 459545757cbfdd6af031143c67a618335ba9f43b[m
+Merge: 72e8738 d6552ab
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Aug 7 13:35:59 2023 +0000
+
+ Merge branch 'vector-unit-tests' into 'develop'
+
+ Add proper unit tests for 'Vector' and its subclasses
+
+ See merge request magnet/declearn/declearn2!55
+
+[33mcommit d6552ab77140ad0a089d7fa58fa014f43d693e7a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 13:53:26 2023 +0200
+
+ Rename 'Model' unit tests files.
+
+[33mcommit 730beced4d9ee93d11ad73b7ae467fb29e88f200[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 13:52:38 2023 +0200
+
+ Rename 'Vector' unit tests files.
+
+[33mcommit 7c5990c200508444575d406de205617d0e9d1dd0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 12:57:59 2023 +0200
+
+ Remove old 'Vector' unit tests.
+
+[33mcommit f5067bdb60cd650ffaad0e9e65852a2b536a2269[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 12:56:45 2023 +0200
+
+ Add some generic and error-cases unit tests for 'Vector'.
+
+[33mcommit 34037cbc1a63d2a23774839db4b7df01b99b50ed[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 12:36:50 2023 +0200
+
+ Add unit tests on (Other)-NumpyVector compatibility.
+
+[33mcommit bcc6903453309b7a85608b207b4145bf013b8d96[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 12:32:54 2023 +0200
+
+ Fix 'Vector.minimum' and '.maximum' type annotations.
+
+[33mcommit 5ccad9c2e09b7827dd24a0e3083c77a3792a2db3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 12:24:11 2023 +0200
+
+ Deprecate 'axis' and 'keepdims' kwargs of 'Vector.sum'.
+
+ These arguments have never been used, are defined at the level of
+ subclasses rather than that of the 'Vector' ABC, and are bound to
+ create some issues if mis-used.
+
+ They are being deprecated, and will be considered for re-instatement
+ with proper ABC definition and documentation if and only if they are
+ required to implement some given algorithm in the future.
+
+[33mcommit 6f811fc00d771fe3d9f8ac4c9d0ed62544cf592d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 11:13:37 2023 +0200
+
+ Fix an error message in 'Vector.build'.
+
+[33mcommit 4395571e90a4ad1d321eae43acb8800ca6d53691[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 11:12:51 2023 +0200
+
+ Add unit tests for 'Vector.build'.
+
+[33mcommit ce6cd7925fe1dbe69a3cd3a11d2a7a9464269d7b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 7 11:00:32 2023 +0200
+
+ Fix type-association for 'JaxNumpyVector'.
+
+[33mcommit de262d7bc515ef8795939f050c8e7a8bcc423a7e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 4 17:08:06 2023 +0200
+
+ Test 'tf.IndexedSlices' support in 'TensorflowVector' unit tests.
+
+[33mcommit 5c21f8596469bef5e913f963669d93eba266f1cb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 4 17:00:26 2023 +0200
+
+ Add square and sqrt Vector unit tests.
+
+[33mcommit 6e1c1bfa92f331fd4ea4047ad6a1865b7abdb704[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 4 15:59:28 2023 +0200
+
+ Fix 'TensorflowVector.shapes' output type.
+
+[33mcommit 6826cfeafd153debf5a30f4a072ba9936aa4f8a6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 4 15:44:20 2023 +0200
+
+ Implement a new unit tests suite for 'Vector' classes.
+
+[33mcommit 72e8738aae686779d945d28dd06252a349558c51[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 4 10:06:05 2023 +0200
+
+ Fix 'SklearnSGDModel' cross-version compatibility.
+
+[33mcommit fadcdc44e66dc74304624cc0ae7d9f944fe30bc3[m
+Merge: 47381cb 14bdb42
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Aug 4 07:28:17 2023 +0000
+
+ Merge branch 'hotfix-truediv' into 'develop'
+
+ Hotfix: 'Vector.__rtruediv__' and 'L2Clipping'
+
+ See merge request magnet/declearn/declearn2!54
+
+[33mcommit 14bdb425f1ac93d3115ffe08c596befa0e73c514[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 3 18:59:36 2023 +0200
+
+ Disable GPU use in 'test_regression' integration test.
+
+[33mcommit ca3dafb7f10e27899e22418073e1a73afe30c92d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 3 18:40:01 2023 +0200
+
+ Deploy '--cpu-only' test option to optimizer tests.
+
+[33mcommit 047d9b64adb16c3e7b84e7252f296914dc71c796[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 3 17:46:41 2023 +0200
+
+ Fix 'L2Clipping' algorithm.
+
+[33mcommit 3a53dfa4e91efc5cf36b44a905be1ab6bcdb55ec[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 3 17:45:39 2023 +0200
+
+ Fix 'Vector.__rtruediv__'.
+
+[33mcommit 47381cb90595b95712aa4aeb538cc1ae1aa10c1f[m
+Merge: 6bb9d42 62669f6
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Aug 1 13:05:04 2023 +0000
+
+ Merge branch 'enhance-datasets' into 'develop'
+
+ Enhance `declearn.dataset`: implement `TensorflowDataset` and improve `TorchDataset` and `InMemoryDataset`.
+
+ Closes #21
+
+ See merge request magnet/declearn/declearn2!53
+
+[33mcommit 62669f67ea457689a25daa44292ac813361bb15c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jul 31 16:34:06 2023 +0200
+
+ Implement and test 'collate_with_padding' util for 'TorchDataset'.
+
+[33mcommit 6c24ad11bdb4baacc33431f80360558c6d78abe4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jul 31 16:33:35 2023 +0200
+
+ Add 'collate_fn' parameter to 'TorchDataset'.
+
+[33mcommit 2581aca71b84b49dec82a9c3572c25108661a53c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jul 31 13:26:24 2023 +0200
+
+ Add support for padded and ragged batching to 'TensorflowDataset'.
+
+[33mcommit aab0c4f82911a3a7dd3a1848b1a69cc4231dac88[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jul 31 10:48:09 2023 +0200
+
+ Implement custom 'PoissonSampler' for 'TorchDataset'.
+
+ Previously, poisson sampling was delegated to a sampler implemented
+ as part of the third-party 'opacus' library. However, the latter is
+ not mandatory for end-users, and its implementation does not prevent
+ empty batches from being yielded (which break our collate function).
+
+ Therefore, this commit implements our own sampler, as it is a very
+ straightforward process.
+
+[33mcommit 9a85bab03b8e6707c85110d9962663a0af0261c6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jul 31 10:47:05 2023 +0200
+
+ Add unit tests for batch-generation with replacement.
+
+[33mcommit 742c8d8763facc6a95188042cb0bf249e16c480f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jul 31 10:13:07 2023 +0200
+
+ Add base unit tests for 'TensorflowDataset'.
+
+[33mcommit 4fc12a667ff2d12ce1c949a137ca40c9e0034881[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 28 18:22:29 2023 +0200
+
+ Implement 'replacement' parameter use for 'InMemoryDataset.generate_batches'.
+
+[33mcommit 09c5fb27f78dd3ec02ce8f77ed332f0b9d750e8b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 28 16:35:00 2023 +0200
+
+ Implement 'TensorflowDataset' interface to 'tf.data.Dataset'.
+
+[33mcommit 6bb9d428c7ba52444f50fa02f82e83e46f739a30[m
+Merge: 1bf2d69 114ecf3
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Jul 28 09:37:57 2023 +0000
+
+ Merge branch '24-declearn-text' into 'develop'
+
+ Implement 'TorchDataset' to support an example on language models
+
+ Closes #24
+
+ See merge request magnet/declearn/declearn2!47
+
+[33mcommit 114ecf35ee075ceb82eb03fdc1990ebe1a5d23c7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 26 23:10:43 2023 +0200
+
+ Fix 'to_numpy' test util.
+
+[33mcommit e7e2880b622fc26fd1b32fb98a30100a016b8a22[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 26 17:40:07 2023 +0200
+
+ Improve 'TorchDataset' setup in 'test_regression'.
+
+[33mcommit c1dbe133c15ff17951a971c667e863e0623b9a8f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 26 17:18:54 2023 +0200
+
+ Use 'make_importable' to perform local imports in tests' code.
+
+[33mcommit db97bed55c473991faf675470443b46d9e43e11b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 26 16:10:55 2023 +0200
+
+ Move 'TorchDataset' to 'declearn.dataset.torch' manual-import submodule.
+
+[33mcommit be4c9f220537b5529b6648d6b42868455608600d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 26 15:57:59 2023 +0200
+
+ Correct 'TorchModel._unpack_batch' output signature.
+
+[33mcommit 67a9a0177598227a76bddd95ccaa4031e22f57b5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 26 15:44:24 2023 +0200
+
+ Fix 'TorchDataset' custom collate fn and its tests.
+
+ - The previous 'transform_batch' function was rewritten into the
+ 'TorchDataset.collate_to_batch' static method. Its dependency
+ util (and a number of 'TorchDataset' attributes) were removed.
+ - The function was revised into yielding expected results so as
+ to collate lists of sample data (formatted as tuples) into a
+ declearn-compatible batch. Its tests were updated and expanded
+ to match real-life requirements.
+
+[33mcommit d94bb87e607147115cf7752436efaa8e5838c448[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 26 15:43:41 2023 +0200
+
+ Remove deprecated '_type_key' attribute from 'Dataset' API.
+
+[33mcommit 6ae7dd7b01b3163fdd231b7fce3047a778da5fe8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 26 15:38:35 2023 +0200
+
+ Marginally improve 'TorchDataset' backend code.
+
+ - Wrap up 'opacus' conditional import to raise proper warnings.
+ - Favor 'torch.utils.data.X' calls rather than calling 'X' after using
+ 'from torch.utils.data import X'.
+ - Catch exceptions (and silence mypy warnings) due to possibly-unsized
+ torch Dataset inputs.
+ - Clean up a few minor development artifacts in the backend code.
+
+[33mcommit f8c33496fd63752e3ae8a3387d3e8b8f53286a03[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Jul 7 11:32:23 2023 +0200
+
+ Making data_item_info an attirbute of TorchDataset
+
+[33mcommit ced87128aa33b61d06fdd98e06c38ee684e2f9da[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Jul 4 18:19:47 2023 +0200
+
+ Centralize all versions of the "to_numpy" function in a single test_util
+
+[33mcommit 768c79d78f45957280839a7c92bd5124f3c7b41d[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Jun 29 17:31:32 2023 +0200
+
+ Apply linter-based and typo-resolving corrections.
+
+[33mcommit 33a2224bb1a6d2288b4c1a3b26235ef1cc7b997d[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Jun 29 17:01:07 2023 +0200
+
+ Add unit tests for 'InMemoryDataset'.
+
+[33mcommit 8d236abd94d7a609ac424e7280586e09994f541f[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Jun 29 16:19:13 2023 +0200
+
+ Finalizing Torch Dataset testsc, and moving key fucntions to test_utils
+
+[33mcommit 18db2fbe22ef3757428f251f9e037ca56b26e423[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Jun 27 17:35:18 2023 +0200
+
+ Basic test suite + application to torchdataset
+
+[33mcommit 4286453bdce50a8dedf5a60ee2aaf0059e9a3318[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Jun 23 13:36:07 2023 +0200
+
+ Revise 'TorchDataset', and use it in 'test_regression'.
+
+[33mcommit cbd665b4ae74491e5be2406f3d3ecb22761d16d1[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Jun 22 18:56:29 2023 +0200
+
+ Updating torch dataset v1 - validate method broken
+
+[33mcommit 90119d689038c1d6f35d0d4417bc6e92893057d5[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Jun 21 16:59:51 2023 +0200
+
+ Adding torch dataset v0 and moving TorchBatch from model to dataset folder
+
+[33mcommit c2f3e875d2e04499e7eac8fb367c79f310987067[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Jun 19 20:36:46 2023 +0200
+
+ Trim 'Dataset' ABC, removing load/save method from the API.
+
+ Also add 'replacement' argument to the 'generate_batches' method.
+
+[33mcommit 1bf2d691b0793502509264cfb40a560d8c1db92b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jul 25 11:37:05 2023 +0200
+
+ Update the website docs on tests and ways to run them.
+
+[33mcommit 1943a468524d998342484f800cb6a16b779b5cdf[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jul 25 11:02:22 2023 +0200
+
+ Add documentation for developers to 'run_tests.sh' script.
+
+[33mcommit 31a69ce46b7f4eee7328ff15ba1ee95244607b11[m
+Merge: 50255e3 67ef8f4
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Jul 24 14:43:18 2023 +0000
+
+ Merge branch 'test-magnet-runner' into 'develop'
+
+ Switch to a Magnet-hosted CI runner
+
+ See merge request magnet/declearn/declearn2!45
+
+[33mcommit 67ef8f4c49e51bdef4d2705bd7f90f315c5afc73[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jul 24 16:30:24 2023 +0200
+
+ Force CI tox env recreation on merge commits to a main branch.
+
+[33mcommit 451cde2da441df6ed9e6142a3d9f4d7ee97a745e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 21 13:26:11 2023 +0200
+
+ Enable torch dynamo error fallback in 'Model' unit tests.
+
+[33mcommit 59f8779d9e8f31cf20f30cabfae6f2c2c1309dbb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 20 16:27:40 2023 +0200
+
+ Change CI/CD runner and execution rules.
+
+ - Move from shared inria CI/CD runners to a Magnet-hosted one, that
+ enables re-using the same docker volume across jobs.
+ - Ensure the venv, pip and tox cache folders are preserved across jobs.
+ - Ensure the tox venvs differ between branches, preserving isolation.
+ - Ensure merged branches' tox venv is garbage-collected.
+ - Enable removing the current branch's cached tox venv.
+ - Enable removing the entire tox cache directory.
+
+[33mcommit b099959bd8e88bb6f7f070d8227e055719d6b008[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu May 4 11:42:14 2023 +0200
+
+ Switch to a Magnet-hosted CI runner.
+
+ - This commit squashes (effectively dropping most of) a number of
+ attempts at tweaking the CI/CD config so as to run on a custom,
+ self-hosted, gpu-enabled CI runner and make an efficient use of
+ caching capabilities in order to reduce CI pipelines' runtime.
+ - Although the finalized commit barely comprises any change it is
+ meant to record that Nathan was the one to first set up our new
+ CI runner and tackle GPU-access issues, and put some effort in
+ attempting to tackle caching issues that would later be handled
+ by Paul.
+
+[33mcommit 50255e378fecb2dc28ff5dc3d164cc3ebc7c3b1f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jul 17 11:27:58 2023 +0200
+
+ Increase CI/CD coverage statistic precision.
+
+[33mcommit 571ff5007911c7dabb23795f9e0c15c44ad6a9c5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri May 26 10:58:59 2023 +0200
+
+ Refactor 'gen_docs.py' code.
+
+[33mcommit 4cff0d377a246bb06fa186a7905692d0aa84980d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 24 13:35:04 2023 +0200
+
+ Refactor 'run_as_processes' backend code.
+
+[33mcommit 861d559b520f8761f1811ad68106633ed08f856c[m
+Merge: 9f0035d 6d3695f
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Thu Jul 13 07:54:27 2023 +0000
+
+ Merge branch '27-add-support-for-torch-2-0' into 'develop'
+
+ Add support for Torch 2.0
+
+ Closes #27
+
+ See merge request magnet/declearn/declearn2!49
+
+[33mcommit 6d3695fdd1e4ad162440ca99179743fc1caa856e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 7 12:05:50 2023 +0200
+
+ Add a couple of 'no cover' pragmas.
+
+[33mcommit fdb8167fc5ecc4c3a68b7564ed43577a9c59ff73[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 13 08:41:09 2023 +0200
+
+ Have some unit tests re-run using Torch-1.13 in the CI.
+
+[33mcommit 7468f705fb2dd4d020daf0586f85f09d3443582c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri May 26 10:44:55 2023 +0200
+
+ Add and document 'torch.compile' support.
+
+ - Torch 2.0 introduced `torch.compile`, a novel utilty to optimize
+ computations via their JIT-compiling into optimized kernels. At
+ the moment, compiled modules cannot be saved (but their states,
+ which are shared with the underlying original module, can), and
+ are not compatible with `torch.func` functional execution either.
+ - As declearn vows to support Torch 2.0, it seems crucial that end-
+ users may use `torch.compile`. However, as long as Torch 1.10-13
+ versions (and previous versions of DecLearn 2.X) are supported,
+ this handling should not break backward compatibility.
+ - An initial approach was to enable compiling the handled module as
+ part of `TorchModel.__init__`. However, this proves impractical,
+ as it takes away the assumption that end-users should be able to
+ use their customly-prepared module as-is - including pre-compiled
+ ones. This is all the more true as their are many options to the
+ `torch.compile` function, that DecLearn has no purpose handling.
+ - Therefore, the approach implemented here is to detect and handle
+ models that were compiled prior to being input into `TorchModel`.
+
+ - Here are a few notes and caveats regarding the current implementation:
+ - Some impractical steps were taken to ensure weights and gradients
+ have the same name regardless of whether the module was compiled
+ or not, _and regardless of the specific 2.x version of declearn_.
+ When we move to declearn 3, it will be worth revising.
+ - A positive consequence of the previous note is that the compilation
+ of the module should not impair cases where some clients are using
+ torch 1 and/or an older 2.x version of declearn.
+ - The will to retain user-defined compilation option is mostly lost
+ due to the current lack of recording of these info by torch. This
+ is however expected to evolve in the future, which should enable
+ sharing instructions with clients. See issue 101107 of pytorch:
+ https://github.com/pytorch/pytorch/issues/101107
+ - A clumsy bugfix was introduced to avoid an issue where the wrapped
+ compiled model would not take weights updates into account when
+ running in evaluation mode. The status of this hack should be kept
+ under close look as the issue I opened to report the bug is treated:
+ https://github.com/pytorch/pytorch/issues/104984
+
+[33mcommit ef8b5662e43533cd59f439f467b1ac08e7a419e3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jul 4 13:43:23 2023 +0200
+
+ Add support for torch model buffers in clipped gradients computation.
+
+[33mcommit 4202d5624820e1a57931757b44bbe546a0653db3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jul 4 17:10:31 2023 +0200
+
+ Update 'pyproject.toml' to support both Torch 1.1X and 2.X.
+
+ - Turn "all" and "torch" extras more loose, enabling (thus, in most cases,
+ resulting in) installing torch >=2.0 rather than 1.1[0-3].
+ - Add "torch1" and "torch2" extras to specify explicitly whether the user
+ would rather use Torch 1.13 or >=2.0. Opacus is made part of these two
+ extras, as its version is also co-dependent with Torch/Functorch.
+ - Document this change as part of the package installation guide.
+
+[33mcommit accb07d1ba8815c3b6fbe659330cb740238dda84[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 24 16:48:21 2023 +0200
+
+ Revise TorchModel to support Torch 2.0.
+
+ - Torch 2.0 was released in march 2023, introducing a number of
+ new features to the torch ecosystem, while being non-breaking
+ wrt past versions on most points.
+ - One of the salient changes is the introduction of `torch.func`,
+ that integrates features previously introduced as part of the
+ `functorch` package, which we have been relying upon in order
+ to efficiently compute and clip sample-wise gradients.
+ - This commit therefore introduces a new backend code branch so
+ as to make use of `torch.func` when `torch~=2.0` is used, and
+ retain the existing `functorch`-based branch when older torch
+ versions are used.
+ - As part of this effort, some of the `TorchModel` backend code
+ was refactored, notably deferring the functional transform of
+ the input `torch.nn.Module` to the first call for sample-wise
+ gradients computation. This has the nice side effect of fixing
+ cases when users want to train a non-functorch-compatible model
+ without DP-SGD.
+
+[33mcommit 9f0035d7d7a80b8a28181c2386154e700681cc27[m
+Merge: 8843e07 4efde98
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Jul 12 13:25:58 2023 +0000
+
+ Merge branch 'revise-ci-and-tests' into 'develop'
+
+ Refactor tests pipeline, introducing 'scripts/run_tests.sh'.
+
+ See merge request magnet/declearn/declearn2!52
+
+[33mcommit 4efde9876c7762d72b67bd7415e3db2315b25dea[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jul 10 13:23:45 2023 +0200
+
+ Adjust the CI/CD configuration to 'tox.ini' evolution.
+
+[33mcommit 31d3fb67f1b254797d48d985326e5fdb7b374120[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 7 12:05:08 2023 +0200
+
+ Refactor tests pipeline, introducing 'scripts/run_tests.sh'.
+
+ - Add the 'scripts/run_tests.sh' bash script, which enables running
+ various blocks of tests. This mostly refactors instructions that
+ used to be part of 'tox.ini', in a more convenient way:
+ - some tests are grouped (commands in a group _all_ run, but the
+ group fails if any has failed once all are done);
+ - instructions are (hopefully) easier to read through and should
+ be easier to maintain in the future;
+ - it should be easier to decline more groups in the future (e.g.
+ to have a routine with only unit tests);
+ - groups of tests can be run easily, including without using tox.
+ - Have the 'tox.ini' file trigger some groups of tests by calling
+ the new bash script. This makes the tox file easier to read and
+ parse through, and merely oriented at providing run isolation.
+ - Split test groups into various envs in tox, which may be slightly
+ more costly to run, but enables writing simpler CI/CD pipelines
+ that merely pass arguments to the tox command.
+ - Make it explicit in 'tox.ini' that tests can be run with python
+ 3.8 and onwards.
+ - Keep a "main" job that groups all others. This is useful in the
+ Inria shared Gitlab Runner, as caching is disabled, so that the
+ use of distinct jobs would result in redundant costful package
+ installations.
+
+[33mcommit 8843e075e5b75e52f03ab0cc35aa961e14936f49[m
+Merge: c9f5245 6b7a3d6
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Jul 12 13:00:15 2023 +0000
+
+ Merge branch 'mnist-example' into 'develop'
+
+ Revise the MNIST example, cleaning up the code and documentation.
+
+ See merge request magnet/declearn/declearn2!51
+
+[33mcommit 6b7a3d618cc67d957d8702672e86c6acd727b38b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jul 11 15:08:00 2023 +0200
+
+ Revise the MNIST example, cleaning up the code and documentation.
+
+[33mcommit c9f52458b8cdc3eab7e89eb57d42faa6b5e4cc3c[m
+Merge: 3b687a3 38accb6
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Thu Jul 6 16:52:11 2023 +0000
+
+ Merge branch 'sklearn-13-dtype-fix' into 'develop'
+
+ Fix compatibility issues with scikit-learn 1.3.0 and tensorflow 2.13.0
+
+ Closes #28 and #29
+
+ See merge request magnet/declearn/declearn2!50
+
+[33mcommit 38accb6466e94f2b68955dfa7692a5511c53fa6d[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Jul 6 17:12:05 2023 +0200
+
+ Refactor 'SklearnSGModel' dtype choice code into a private function.
+
+ Co-authored-by: Paul ANDREY <paul.andrey@inria.fr>
+
+[33mcommit c4fe82b50fcddfed71c6701db8c6c8c69570a748[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 6 14:13:16 2023 +0200
+
+ Prune 'test_main.py' pytest-collected test scenarios.
+
+ The legacy 'test_main.py' integration tests are now mostly redundant
+ with other integration tests, adding little marginal value to the CI/CD
+ tests pipeline. From a Coverage POV, the only thing that is not already
+ tested elsewhere is the actual use of Scaffold.
+
+ Hence the test scenarios collected during PyTest runs are restricted to:
+ - FedAvg and Scaffold variants of Optimization strategies.
+ (prior: FedAvg + FedAvgM + Scaffold)
+ - In non-fulltest, only the Scaffold strategy is used.
+ (prior: FedAvg + Scaffold)
+
+ This results in testing 3 scenarios in the limited pipeline and 18 in the
+ "fulltest" one, when there previously would be 6 or 27 scenarios.
+
+[33mcommit 29ecf47396bccd1baf1363d1f3806bfa083d1c61[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 6 14:11:37 2023 +0200
+
+ Add 'dtype' use to 'SklearnSGDModel' unit tests.
+
+[33mcommit 95cfa2b9069472f2867120d099d330a88e6d166f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 6 14:09:39 2023 +0200
+
+ Refactor and fix 'build_keras_loss' util; add unit tests for it.
+
+[33mcommit 808cd5fd0a9a0dc3a25fc8b6d6f10b34341520dc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 6 11:56:48 2023 +0200
+
+ Fix 'build_keras_loss' for TensorFlow 2.13.
+
+[33mcommit 02afef6ec467896cc9fdcdaab8df238860e40ef1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 5 18:07:21 2023 +0200
+
+ Improve dtype handling in 'SklearnSGDModel'.
+
+ - Scikit-Learn 1.3.0 introduced the possibility to use any dtype for
+ SGD-based models' weights.
+ - As a consequence, this commit introduces the optional 'dtype' argument
+ to 'SklearnSGDModel.__init__' and '.from_parameters' methods, which is
+ only used with `sickit-learn >=1.3` and an un-initialized model.
+ - Coherently (and for all scikit-learn versions), additional type/dtype
+ verification has been implemented under the hood of `_unpack_batch`.
+
+[33mcommit 3b687a3a249c7f33711da02f4ab575ff151869ff[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jun 8 14:23:35 2023 +0200
+
+ Fix Heart Disease UCI Dataset loaded due to website change.
+
+[33mcommit afc295100261d5a19b278363e73763eafc2022ba[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 31 11:57:23 2023 +0200
+
+ Replace deprecated 'pkg_resources' with 'importlib.metadata'.
+
+[33mcommit a8cfa7cc9fa9045d13466527793f8baf706a817c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue May 23 16:22:15 2023 +0200
+
+ Remove unused line in 'tox.ini'.
+
+[33mcommit f7a13dff9ce3de674ac263628ae8dc98153b2335[m[33m ([m[1;33mtag: v2.2.0[m[33m)[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu May 11 15:58:17 2023 +0200
+
+ Fix Jax dependency specification.
+
+[33mcommit 9113e8adfca5678b20e33f8eb6b58f8135c14125[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu May 11 15:53:42 2023 +0200
+
+ Document known issues with GPU support.
+
+[33mcommit 88a990ec327925d7194c283169f7e054af92c80b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu May 11 10:39:47 2023 +0200
+
+ Turn device-selection 'UserWarning' into 'RuntimeWarning'.
+
+[33mcommit 9d1b25cb591e4713f303b6441892eff5faf1f8bd[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu May 11 10:39:10 2023 +0200
+
+ Fix device selection for Jax and add '--cpu-only' to its tests.
+
+ - Fix device-selection utils for Jax/Haiku.
+ - Ensure any GPU is left untouched when the device policy specifies
+ that the CPU must be used.
+ - Fix instructions in both HaikuModel and its tests' setup code to
+ achieve the former.
+ - Note: `jax.jit(..., device=<D>)` is used to prevent Haiku from
+ using another device than the desired one; however this argument
+ is being deprecated, making this solution short-lived. Since the
+ supported Jax version is hard-coded however, we keep this patch
+ for now, in wait of future framework evolutions.
+ - Add the use of the optional, custom `--cpu-only` flag in Haiku
+ unit tests. This enabled testing the former changes in an env
+ that has a detectable GPU, but an unproper driver version (so
+ that any undesired use of that GPU would make the tests fail).
+
+[33mcommit 28e881522f3f4da1d4ff7542065f6063893fb3c2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 10 15:10:30 2023 +0200
+
+ Bump version number to 2.2.0.
+
+[33mcommit 1eae849db8486fef14b5fb7304d2bea8b71e9a3b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 10 15:05:22 2023 +0200
+
+ Improve some documentation formatting.
+
+[33mcommit 164ddb3b4b827c2fbd3ed047be4fee339adcdcbe[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 10 15:03:18 2023 +0200
+
+ Improve 'HaikuModel.set_trainable_weights' docs.
+
+[33mcommit dd2eef5029ab0a75785372326d5e3fe472c72799[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 10 14:49:12 2023 +0200
+
+ Fix jax dependency specification.
+
+[33mcommit e95ece2ee68e98a570539c1f3eac4d68d83e571e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 10 14:48:56 2023 +0200
+
+ Fix 'declearn.model' main module docstring.
+
+[33mcommit 38dbe123376bc69097098fc7406e5f1f5ab69860[m
+Merge: a748ae7 eef2033
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed May 10 09:30:49 2023 +0000
+
+ Merge branch 'gardening' into 'develop'
+
+ Minor gardening around the package
+
+ See merge request magnet/declearn/declearn2!44
+
+[33mcommit eef2033fc3d4ceda652cdccf863738312a434b85[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 10 09:53:43 2023 +0200
+
+ Enhance type hints for Jax/Haiku classes.
+
+[33mcommit ba2caa2f9b694283f1283266ce4a386885310af1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 3 16:40:38 2023 +0200
+
+ Revise test suite's ordering.
+
+ - First run unit tests (which are the more problematic).
+ - Then run code quality verifications (pylint / mypy / black).
+ - Finally run integration tests.
+
+ The rationale to inverting the last two items relative to what was
+ done prior to this update is to keep the longest-running tests for
+ the end (rather than re-play a long pipeline due to late-detected
+ formatting issues).
+
+[33mcommit 9e6f9a8c6ad0e6472fa450cd29c2a5279063e7ed[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 3 16:40:12 2023 +0200
+
+ Patch 'TomlConfig' unit tests for PyLint.
+
+[33mcommit fd8abbd3f7b27cba1087f441f32fd734dc8ef836[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 3 15:57:10 2023 +0200
+
+ Add the 'list_aggregators' util function.
+
+[33mcommit c3a6186149d09fc0d3d1c0ec74058f109cedf375[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 3 10:57:10 2023 +0200
+
+ Add some 'pragma: no cover' comments.
+
+[33mcommit d37d8ed21548984bdfb9112d676d8bb469cc8ab5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 3 10:46:17 2023 +0200
+
+ Lift 'run_as_processes' backend code.
+
+[33mcommit 93f13d536ed0dd919be456f0a3adf0380eefb088[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 3 10:40:42 2023 +0200
+
+ Add basic unit tests for 'Aggregator' classes.
+
+[33mcommit 29c6fb3a0cfc7f73e528d4db1a57dd9d01c6f2d8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed May 3 10:40:07 2023 +0200
+
+ Marginally improve some 'Aggregator' code and docs.
+
+[33mcommit f9ee9b9a3c19829970dca24d8e3856f699856bb2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Apr 28 11:48:30 2023 +0200
+
+ Add minimal functional test for quickrun mode.
+
+[33mcommit 163acc1f64e462f56d06abc1c704ab2baae311ec[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Apr 28 11:47:16 2023 +0200
+
+ Make quickrun mode compatible with gRPC.
+
+[33mcommit fd7d8c2a56609babd92ff5c93c228cdeddfd93e4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Apr 28 11:46:52 2023 +0200
+
+ Fix default aggregator in `FLOptimConfig`.
+
+[33mcommit dd17cd64d10bae0c6a75a49c698c7a3e1eba0d43[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Apr 27 16:38:31 2023 +0200
+
+ Add unit tests for 'TomlConfig' and revise its backend.
+
+[33mcommit 2b020369a21697513204a312a1d4dc7a7c53220d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Apr 27 10:42:40 2023 +0200
+
+ Extend unit tests for type-registration utils.
+
+[33mcommit 6630eb915dfd8c59304f466fde9f01efd5250574[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 17:14:46 2023 +0200
+
+ Add utils to list available optimizer plug-ins.
+
+[33mcommit 13b5c75792879db557e45e23900624c5ce6e17eb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 16:45:56 2023 +0200
+
+ Add 'Model.get_wrapped_model' to the API.
+
+[33mcommit 7d440b0cf60b28cb5875b20ec6a1ab2ffd0c8cf3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 16:25:16 2023 +0200
+
+ Correct a comment in 'NoiseModule' code.
+
+[33mcommit 7f3327b1bc94309fc4babc06ef7c6a5027986be6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 16:18:50 2023 +0200
+
+ Enhance typing of 'Model' and 'Vector' using 'typing.Generic'.
+
+ - Make `Vector` a `Generic[T]` to indicate that stored data arrays
+ should have coherent types.
+ - Make `Model` a `Generic[VectorT]` to indicate that a given model
+ class is associated with a unique vector class.
+ - In `Optimizer` and plug-ins ABCs, specify that inputs and outputs
+ should be coherent using `Vector[T]` and `Model[Vector[T]]` type-
+ hints.
+ - In concrete plug-in classes, leave the existing code as-is so as
+ not to make it harder to read.
+
+[33mcommit a748ae7131fbd0be50aa947f146e78e44b66da95[m
+Merge: 2258d64 11274f4
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue May 9 19:16:39 2023 +0000
+
+ Merge branch 'haiku-jax' into 'develop'
+
+ Add support for Jax / Haiku
+
+ See merge request magnet/declearn/declearn2!32
+
+[33mcommit 11274f419a6b35b9cedfc4f9f0833da6c39f9729[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue May 9 14:40:27 2023 +0200
+
+ Removing unused typing import
+
+[33mcommit 6653b3dc55959414ce800e3281f9476d9c7e2653[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue May 9 12:55:47 2023 +0200
+
+ Avoid re-implementing haiku model traverse.
+
+[33mcommit 5f33eef6bd3ffd819a713c6e7d99110d6eeb26c0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri May 5 12:16:33 2023 +0200
+
+ Revise 'HaikuModel' backend.
+
+ * Add proper weights and gradients naming.
+ * Store the model's nested dict of parameters rather than its
+ flattened counterpart.
+ * Refactor some backend code to improve readability of gradients
+ computation.
+ * Add a public getter to access weights' names, opt. restricted
+ to trainable ones.
+ * Revise `set_trainable_weights` to received lists of names rather
+ than indices. Keep support for predicates and PyTree inputs.
+
+[33mcommit d02b880ed7c59aab0ffda0519f1cef2440103da5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri May 5 10:20:55 2023 +0200
+
+ Minor formatting revisions to 'HaikuModel'.
+
+[33mcommit defe9ba664f9c0e4ea4de5ca37bd1701c1f255e3[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed May 3 17:53:04 2023 +0200
+
+ post rebase corrections
+
+[33mcommit 692a4054753c0deaab7ef406d1b20e96a3fc102a[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Apr 24 19:30:03 2023 +0200
+
+ linting
+
+[33mcommit eb441268a48f093bfb489050de12984494c2a7a5[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Apr 24 19:24:30 2023 +0200
+
+ Adding list inputs to model and automatic unpacking
+
+[33mcommit d35056847ba5d3e7324da6556f9470f43b36163b[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Apr 24 18:06:12 2023 +0200
+
+ Changign test model to use last LSTM hidden state
+
+[33mcommit a55f6417aa56f5fd027fdd4fdbfd1e335964507b[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Apr 24 17:49:46 2023 +0200
+
+ Fixing `_compute_batched_gradients_clipped` test and typos
+
+[33mcommit 98090019afb4ce68cc131c4383ca538513a4b845[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Apr 24 13:18:36 2023 +0200
+
+ Fixing Jax RNN `_compute_batch_gradient` testing - with open questions on the internal logic of the model
+
+[33mcommit 4bea61f0b1be169c49937100b09ea5db71f99db9[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Apr 19 18:26:27 2023 +0200
+
+ Running gradient only on trainable parameters
+
+[33mcommit c948f95a5f99a2ded442972db280a44b88565a40[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Apr 19 17:20:24 2023 +0200
+
+ Refining haiku model frozen weights methods
+
+[33mcommit fb6cc0b42e5eef3843f142bcb8d2f248646afd37[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Apr 19 17:19:19 2023 +0200
+
+ Adding frozen weights tests
+
+[33mcommit bac15a39e34c7a2d79fd21b6a7c3935b10944315[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Apr 18 19:14:49 2023 +0200
+
+ v0 of adding support for frozen_weights. Main ideas:
+ * Store the index of trainable weights under `self._trainable`
+ * Create a `set_trainable` method to allow to define trainable weights by using either a list of param leaves index, an existing pytree or dictionnary, or a boolean function checking for conditions on the param pytree
+ * Only update what the weights needed
+
+[33mcommit 1d43c0ddff3cd6283400f1ba76d5604bf760bb6b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 17 10:04:54 2023 +0100
+
+ Marginal 'HaikuModel' changes after review.
+
+[33mcommit 28a34851693f5a2cb220acd1c8eea4240624d434[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 17:46:49 2023 +0100
+
+ Silence duplicate-code issue and minify haiku test code.
+
+[33mcommit e1031b5ede39043471dd5ef54960008eb5c6e3a2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 17:15:49 2023 +0100
+
+ Add missing license headers to haiku-related files.
+
+[33mcommit 56971bd5932d0c3cb972757e2417b8d0bcddacfe[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 17:03:00 2023 +0100
+
+ Improve some Haiku/Jax related documentation and imports.
+
+[33mcommit 569106f8b29db6064a804b0dce5f49f5223cae27[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 17:02:17 2023 +0100
+
+ Remove redundant jax config call.
+
+[33mcommit 8bcbc9d59de34b7ddf4ce24a9c316665f81bb819[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 16:30:29 2023 +0100
+
+ Clean up 'JaxNumpyVector' backend code.
+
+[33mcommit 0c43b027a5ce89fd0fca0645639871ff18fe7026[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 16:27:56 2023 +0100
+
+ Capture expected warnings in 'HaikuModel' tests.
+
+[33mcommit 1ede1179fae64ef67673be4b22d7ff18c121e549[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 16:12:19 2023 +0100
+
+ Correct Haiku test CNN parameters.
+
+[33mcommit 8dfc07d04be402a05501e8cb85aac62c86727e0d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 15:49:07 2023 +0100
+
+ Correct gradient-clipping with 'HaikuModel'.
+
+[33mcommit 274ae23f6d2eea5c0dfe186dbe55630ee80b58de[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 15:20:12 2023 +0100
+
+ Optimize `HaikuModel` backend.
+
+ Avoid building the gradient-computing jax functions every time they
+ are to be used. On the `test_regression.py` functional test, this
+ results in halving the execution time (which contains that of the
+ sklearn benchmark; hence the actual haiku gain is even higher).
+
+[33mcommit f8330bcefb2256525c7c47975c949e33ed09d44d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 13:10:32 2023 +0100
+
+ Improve 'HaikuModel' backend code readability.
+
+ * Remove some targetted imports to make jax functions' origin
+ clearer when used in the declearn code.
+ * Use `haiku.Params` type-hint to annotate parameters.
+ * Rename the parameters-holding private attributes to make them
+ shorter.
+ * Improve the type-hinting of the parameters-holding attributes.
+ * Add some assertions to silent mypy. These seem un-hurtful as to
+ code runtime and may be skipped with the proper python flag, but
+ may be removed in favor of restoring `type: ignore` comments if
+ they turn out to be costly.
+
+[33mcommit c859b4d7b13d0d60527350973c17dd8f8b6fd609[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 12:30:58 2023 +0100
+
+ Fix and silent pylint warnings.
+
+[33mcommit 1fddb802e3a6a9a018311993cf4f60afc148a2d3[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Mar 15 14:41:56 2023 +0100
+
+ Adding cpu flag in jax install - to be revised when adding gpu support in ci/cf
+
+[33mcommit 5158a84ed9e058ed2c517ebef15b6335772495c1[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Mar 14 17:57:47 2023 +0100
+
+ Make imports conditional on framework availability in test_regression
+
+[33mcommit 0b7f98c7301289895fc42cf92c9338a8b582231a[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Mar 14 17:46:59 2023 +0100
+
+ Adding dependencies in toml, updating jax and haiku version, implementing new typing using jax latest features
+
+[33mcommit b26e1b0cb7e885d393fec13f99a8e8c99b1e6cd6[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 13 19:11:13 2023 +0100
+
+ Adding doc
+
+[33mcommit 2bef2b8146255845390756a1fc64bab2d8faf25f[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 13 18:16:28 2023 +0100
+
+ Removing temp test file
+
+[33mcommit 7d70503ea46d08d008f1d4eb547cf2ba9029f996[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 13 18:15:23 2023 +0100
+
+ jax functional testing
+
+[33mcommit f53e80c77006618caceeeef417223d9e80227df2[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 13 12:39:02 2023 +0100
+
+ Adding vector doc and applying left-most device placement rules
+
+[33mcommit a7d9b3c039302f95236b8656bbe46cb4714c3eb3[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 13 12:16:00 2023 +0100
+
+ Correcting test model shape
+
+[33mcommit f9daded96d9925d064f885382c2a23fb02d4042c[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 13 12:09:44 2023 +0100
+
+ Correcting tf file typos
+
+[33mcommit 6227db0af4d1fd6eff869a608bb59c9471bf5de7[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Mar 10 18:14:44 2023 +0100
+
+ HaikuModel Unit testing
+
+[33mcommit 10399088fb3c2ec341dd610d6fdd80f845123901[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 9 15:30:30 2023 +0100
+
+ Adding jax to testing and correcting JaxNumpyVector for tests
+
+[33mcommit 00e8763b7077ec0e66ae8f578dcb5fdfa59311c1[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 9 14:27:41 2023 +0100
+
+ Adding data_info attribute to HaikuModel
+
+[33mcommit 83ff7106655a82cde4afbdc955b8ce9e936ffe86[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Mar 8 19:42:39 2023 +0100
+
+ Adding feature_shape (WIP)
+
+[33mcommit a709875bbfe2cafd7792eaddc12c96b0589a97a1[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Mar 8 16:42:07 2023 +0100
+
+ linting code
+
+[33mcommit 4fbf74fe8ea023355540f94629f03fd71bbfd186[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Mar 7 20:18:51 2023 +0100
+
+ Adding first simple typing
+
+[33mcommit 9c9943d9fefae86c8342d78ac78748bd20958222[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Mar 7 19:06:43 2023 +0100
+
+ Correcting rnadom generation
+
+[33mcommit d55ec54ad6e401f408d05cf5b647895914a5488f[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 6 10:21:00 2023 +0100
+
+ quick and dirty functional test
+
+[33mcommit deba84259380474ea5ae19c20f3f50c9be0cec88[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 6 10:20:39 2023 +0100
+
+ temp data info unpacking
+
+[33mcommit 89196050dba0e8477b6a3be4c48bfaafd2ee58c7[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 2 19:16:46 2023 +0100
+
+ v0 of GPU support for HaikuModel - untested
+
+[33mcommit 2af9a079c7bc0ea5d014b3f69ceab1d0658d70f6[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Feb 28 18:32:57 2023 +0100
+
+ Adding @jit for grad and vmap operations
+
+[33mcommit 90da8fb30c383de616b8153aa37e0246a232d5a1[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Feb 27 19:25:48 2023 +0100
+
+ Adding functional clipping
+
+[33mcommit 2788d8e686c88b58acf28117349bb21a734d76bb[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Feb 16 18:35:07 2023 +0100
+
+ First functional HaikuModel
+
+[33mcommit 60eaa705e7c783109ebf286e2ab8ff4e04a44283[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Feb 15 14:01:06 2023 +0100
+
+ __init__ files
+
+[33mcommit c1f7b1610546a1843622469107de43719acb927a[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Feb 15 14:00:47 2023 +0100
+
+ Minimal, untested jax numpy vector
+
+[33mcommit 2258d64f3b79136e39785fea75b4973fcedb751e[m
+Merge: 16ec423 365c980
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Apr 26 10:02:18 2023 +0000
+
+ Merge branch 'experimental' into 'develop'
+
+ Quickrun mode
+
+ See merge request magnet/declearn/declearn2!41
+
+[33mcommit 365c9808534be602c724039f4b791ff469da86b5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 11:02:58 2023 +0200
+
+ Fix formatting and links of 'quickstart.md'.
+
+[33mcommit f263e015766bba519de07ee748d91c32b8b7928c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 11:02:44 2023 +0200
+
+ Minor addition to 'setup.md' docs.
+
+[33mcommit 5c2b28a188a844b1ab6ccc99a8945ba4d4cde8e8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 11:02:12 2023 +0200
+
+ Enhance the in-code docs of quickrun.
+
+[33mcommit dba91294fc7ca8b3aebfda79a310913b120fc913[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 10:47:36 2023 +0200
+
+ Expose quickrun building bricks, to render their docs.
+
+[33mcommit 21bcd2f63ca288225398be6a3b7fba1fe3375647[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 10:47:11 2023 +0200
+
+ Fix some docs under 'declearn/dataset/'.
+
+[33mcommit 82174e1eed2766f3530ba9b89df8f1393094e48e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 26 09:35:52 2023 +0200
+
+ Update mnist-quickrun example and add a readme file.
+
+[33mcommit 30549fa951edda572ccd135cb834385a07418978[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Apr 25 18:04:25 2023 +0200
+
+ Change default model name in quickrun mode.
+
+ Rationale: the model must be an instance, while "MyModel" has the
+ naming convention for a class.
+
+[33mcommit 63218f595d18c78a5bb664fb0f2dc766c667fecb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Apr 25 18:03:36 2023 +0200
+
+ Lift the MNIST example files.
+
+[33mcommit a9dd17ded42d465a7e65cabe8daeaf5d15cc3d72[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Apr 25 17:48:52 2023 +0200
+
+ Lift the backend of 'declearn-quickrun'.
+
+[33mcommit c187016b160f8c185566ba055d84aa4d120a5743[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Apr 25 16:27:58 2023 +0200
+
+ Lift up `declearn-split` backend code.
+
+ - Add support for sparse data arrays (as inputs and outputs).
+ - Add support for column name targets when using a csv file.
+ - Make 'split_data.py' file private and expand its documentation
+ as part of `declearn/dataset/__init__.py`.
+
+[33mcommit 648b3fd92f5bb4bb60a3448c838917236464b371[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Apr 25 11:31:05 2023 +0200
+
+ Add support for sparse inputs in 'split_multi_classif_dataset'.
+
+[33mcommit ab6d9c2bb3817dc26a03a8d5e81eef8cb4ef746e[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Apr 21 12:06:14 2023 +0200
+
+ Small formatting changes to quickstart doc
+
+[33mcommit fec24b2c1abdadfa4e080fc0d432435b203d4da1[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Apr 17 17:29:45 2023 +0200
+
+ Minor mypy changes in 'datasets.examples'.
+
+[33mcommit 7f51399d43d05ed966b2e7e160ec128a4bcc60fb[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Apr 14 12:30:56 2023 +0200
+
+ Infering client names form folder names
+
+[33mcommit 8ac8411a47e97e72c9a639747a8af5e6bec62c1c[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Apr 14 11:38:06 2023 +0200
+
+ Local MNIST example
+
+[33mcommit 75c1ec354a78932857aa25e652d0b063d198633f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Apr 13 12:07:35 2023 +0200
+
+ Add unit tests for sparse-data utils.
+
+[33mcommit 69c99d2e7ff3cab402d974345a652fe96296dd5f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 12 16:15:17 2023 +0200
+
+ Add unit tests for data save/load utils.
+
+[33mcommit ab42d2225d3fe25c8819dcb281676d073ac4b0b9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 12 16:13:38 2023 +0200
+
+ Fix bug in 'declearn.dataset.utils.sparse_from_file'.
+
+[33mcommit fc311fd2739236bf4a9458d64fb1db60e0ddfbb8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 12 14:37:33 2023 +0200
+
+ Implement tests for example dataset loaders.
+
+[33mcommit c4b4f80ebeaa89d958fd4e435b82faebab173c8f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Apr 12 14:09:56 2023 +0200
+
+ Remove inadvertantly restored 'test_routines.py'
+
+[33mcommit fe3f18a731d60cfa58e01126af046c2a420d3a58[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Apr 12 12:37:52 2023 +0200
+
+ Linting
+
+[33mcommit 07942a1160400cf5664fb216a2d25ab23db3a0a3[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Mar 31 18:19:02 2023 +0200
+
+ Integrating feedback from Magnet au Lac.
+
+ * Separating `split_data` and `quickrun`
+ * Improving doc and leveraging better existing doc
+ * Using fire for parsing
+
+[33mcommit 3ba1c1aaa476ff07b4064da7dd2943910b35ee31[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 31 16:20:05 2023 +0200
+
+ Fix logger use in quickrun mode.
+
+[33mcommit f398148d6c29cf544a3c562fea7531eb339e8599[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 31 16:16:58 2023 +0200
+
+ Disable GPU use in quickrun mode.
+
+[33mcommit 17d11a023da6f707257aa1cfc9206dbdf7ec3933[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 31 16:16:40 2023 +0200
+
+ Improve file paths handling in quickrun script.
+
+[33mcommit 303b90581c1db988ef15724a4382bd5d846abf13[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Mar 31 12:36:13 2023 +0200
+
+ Heavily simplifying model
+
+[33mcommit 68d3c24285c921ff7f0fc9698c2eee92755defbc[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Mar 31 12:35:57 2023 +0200
+
+ Adding datime to result folder by default
+
+[33mcommit b62f7ed6fbd50dba4d0813e35e9a801d21a8adbb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 31 15:42:17 2023 +0200
+
+ Add a custom logging level to filter multiprocessing outputs.
+
+[33mcommit 041588db2052b60c8c461699a133183b50f56247[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 31 11:44:30 2023 +0200
+
+ Add 'requests' dependency to 'pyproject.toml'.
+
+[33mcommit 838111469023f67519d2b9174499605c10bf38d1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 31 11:31:28 2023 +0200
+
+ Add 'declearn.dataset.utils.split_multi_classif_dataset'.
+
+[33mcommit 8b38b10f0dc536217ef7239788cdaf370998e0f2[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Mar 31 11:29:21 2023 +0200
+
+ Documenting in detail the quickrun mode, later overridden
+
+[33mcommit 4d713710961db07ead1e8db4da6e675a6cca7f64[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 31 10:54:20 2023 +0200
+
+ Moving data_split utils to `declearn.dataset`.
+
+ - Move functional data loading/saving utils to it.
+ - Properly deprecate `InMemoryDataset.(load|save)_data_array`.
+ - Improve docstrings formatting and exhaustiveness.
+
+[33mcommit deb44655135a51c48dec56e3b2adee34a32f8b7a[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 30 21:00:11 2023 +0200
+
+ Small quickrun back end changes
+
+[33mcommit 69699d65e34cc1098e8bb3e67db41bd262a69f82[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 30 20:59:24 2023 +0200
+
+ Simplyfing defualt config and model
+
+[33mcommit 35b5a3af45d71174c26f58c7092c4cf9e1550df9[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 30 20:58:40 2023 +0200
+
+ Adding quicklearn readme
+
+[33mcommit 4206cca1a7011512f84fe8e83c11069de51d6b29[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 30 15:31:31 2023 +0200
+
+ Improve quickrun metrics parsing.
+
+[33mcommit 0296417ea65ad49f9c19757b505fe77859d2f000[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 30 10:29:22 2023 +0200
+
+ Update 'utils.__init__' to document 'run_as_processes'.
+
+[33mcommit 8ddc1bb97015718c5a7d1b344fe0263bb16cab96[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 30 10:27:36 2023 +0200
+
+ Enhance refactored 'run_as_processes' code.
+
+[33mcommit 1c26f2c201bf62a60902cfa9dc74b0c290982911[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 30 09:26:31 2023 +0200
+
+ Adding checkpointer and metrics, re-shuffling data in labels split
+
+[33mcommit 98027bc4ff4ed17e9206c00e97953ebe0218b825[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Mar 29 16:01:08 2023 +0200
+
+ In depth review of TOML parsing, first pass :
+ * Creating "ModelConfig","DataSplitConfig", and "ExperimentConfig"
+ * Relocating and updating parser to accept toml file as input
+ * Update split data to use config input
+ * Modularization and updating of run.py
+ * Small changes to MNIST toml example
+
+[33mcommit ca9f503e8517a5eaac3601342db4939cdf2994ec[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 27 22:55:56 2023 +0200
+
+ Making the MNIST example work
+
+[33mcommit dce90357596c12fb70f9105fadf27ac2d0d491e4[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 27 11:20:58 2023 +0200
+
+ Updating multiprocessing to run on all systems
+
+[33mcommit 22b0af35b38700fcbc5a5a3945d4ba5e2323c207[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Mar 24 12:12:38 2023 +0100
+
+ Debugging and documenting
+
+[33mcommit 0969d20bdbe2ed120be77f53f989bc42d3a71d9a[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Mar 24 11:03:39 2023 +0100
+
+ Adding quickrun cli script
+
+[33mcommit 38f46c0715836c1e37b831cea74f08a6a5b157cb[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 23 12:28:37 2023 +0100
+
+ removing csv export in split data
+
+[33mcommit 8f2be5b7c608efcd570efc9a310bfb4ae21ca9e2[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Mar 22 16:10:45 2023 +0100
+
+ Removing data.py from the example folder
+
+[33mcommit 0341551755bda4e4e11b63c7a5d5434808c3c15b[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Mar 22 16:10:08 2023 +0100
+
+ Adding optional behavior on loading config from toml
+
+[33mcommit 6525cf7663490f040fdf927095d7c033a68efb0c[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 16 09:51:35 2023 +0100
+
+ v0 of quickrun
+
+[33mcommit 16ec423645eae786022598ff43eb8fef3d484be8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Apr 24 15:42:15 2023 +0200
+
+ Silence mypy false-positives on pandas.
+
+[33mcommit 3f6ed06df031255d3699c8d85835af0eb0990ac1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Apr 24 15:17:48 2023 +0200
+
+ Correct RNN test case for Torch.
+
+ - Previously, the predictions would come from the final hidden state
+ of the trained LSTM, rather than the output representation of the
+ final sequence element.
+ - This barely has any consequence given the nature of the unit tests,
+ but improves readability when used as a template for RNN models or
+ similar tests with other model frameworks.
+
+[33mcommit b8222c84df9615dd19220f7a21f11895414ec49c[m
+Merge: 594dfac be449f9
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Apr 17 13:18:49 2023 +0000
+
+ Merge branch 'functorch' into 'develop'
+
+ Refactor 'TorchModel' backend code to compute sample-wise gradients.
+
+ See merge request magnet/declearn/declearn2!42
+
+[33mcommit 594dfacb20e4eca7bbb11c02684241bf1dd0bf5d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Apr 6 11:04:32 2023 +0200
+
+ Minor optimization to 'SklearnSGDModel' backend.
+
+[33mcommit be449f942e0869c899e5fa345d4e522e9d1cb4a6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 17 13:12:56 2023 +0100
+
+ Refactor 'TorchModel' backend code to compute sample-wise gradients.
+
+ * Cache the functorch-built vmapped functions that compute (and opt.
+ clip) sample-wise gradients.
+ * When available, use `functorch.compile` to try and optimize these
+ functions (resulting in lower execution runtimes on repeated calls
+ thanks to their caching).
+ * Note: with the release of torch-2.0, the `functorch.compile` API,
+ which is experimental as of torch-1.13, is already deprecated.
+ Hence this patch is going to be short-lived - but it may in turn
+ pave the way to the future backend changes to add support for and
+ take advantage of new torch-2.0 features.
+
+[33mcommit 46d7492c720ae40c45b6ecbf576fcca4c8208a8f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 29 16:21:38 2023 +0200
+
+ Fix link to API reference page in README file.
+
+[33mcommit a918f2a44db1c1e6017ba73f932306a91d63e9b7[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Mar 29 09:08:16 2023 +0000
+
+ Add coverage statistic capture to CI/CD config.
+
+[33mcommit fe92463734bfa05e9d79a3d272a64758cd9ae3a7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 29 09:49:23 2023 +0200
+
+ Make CI/CD optional 'test-full' non-blocking on main branches.
+
+[33mcommit 1bf6a11bd60f5e0b8c79d98238be4500a4ae86f1[m
+Merge: 34779af d6fe02b
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Mar 28 13:16:43 2023 +0000
+
+ Merge branch 'coverage' into 'develop'
+
+ Compute coverage as part of CI/CD.
+
+ See merge request magnet/declearn/declearn2!38
+
+[33mcommit d6fe02b426ec8e4cf4699692080c278d2a111ea9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 3 12:48:15 2023 +0100
+
+ Add coverage collection to GitLab CI/CD.
+
+[33mcommit a5eedd4ab2afba621f124b95c1d694948aa4326c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 3 12:43:59 2023 +0100
+
+ Add coverage computation to 'tox.ini'.
+
+[33mcommit c8e04ad0981d7a70c82626c89cad74c31c878886[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 2 17:02:08 2023 +0100
+
+ Configure coverage tool with 'pyproject.toml'.
+
+ * Add 'pytest-cov' (hence, indirectly, 'coverage') to third-party
+ dependencies of test environments.
+ * Add configuration of 'coverage' to properly target declearn code
+ (excluding auto-generated protobuf code as well as `test_utils`
+ submodule) and be compatible with the use of multiprocessing in
+ some functional~integration tests.
+
+[33mcommit 34779af93b74e73a964c251ab5bee7d6552ae951[m
+Merge: b744ca4 7d08a3c
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Mar 28 12:34:04 2023 +0000
+
+ Merge branch 'tests-gardening' into 'develop'
+
+ Improve the test suite
+
+ See merge request magnet/declearn/declearn2!39
+
+[33mcommit b744ca4527c14af2bd5355851fd58aa404b29100[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 27 12:03:22 2023 +0200
+
+ Improve API reference navigation in online docs.
+
+[33mcommit 08aea92bccb33b066eb555ac8546543017bbba58[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 27 12:02:52 2023 +0200
+
+ Update gitlab-exported 'docs/index.md' file.
+
+[33mcommit e2643ad1fb6d621bc9954efb89aad02ea947c33d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 27 08:01:32 2023 +0200
+
+ Fix formatting of 'Optimizer' main docstring.
+
+[33mcommit fec1780b3323ad915dbd72dcffb157dc3952788f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 24 21:26:29 2023 +0100
+
+ Update links to the public website.
+
+[33mcommit ae6ec5cd24030b1ec593bb0b35ec3bf65b72131c[m
+Merge: 0664076 cec394f
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Mar 24 14:47:48 2023 +0000
+
+ Merge branch 'docs' into 'develop'
+
+ Restructure the documentation and render docstrings
+
+ See merge request magnet/declearn/declearn2!40
+
+[33mcommit cec394fba05294564a9be273b506dc587571a34d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 24 15:30:54 2023 +0100
+
+ Fix Quickstart docs formatting.
+
+[33mcommit bd89ba1763a8ee8bb009abc6e50ae4785166be00[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 24 15:25:16 2023 +0100
+
+ Remove non-existent website link.
+
+[33mcommit 76a08a32e5b601aba109781c16e176d48e2fbf06[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 24 12:02:38 2023 +0100
+
+ Add docs versioning tools.
+
+[33mcommit 23c4a34b397e325da3e54d8ebd3ebd9dd4751b4b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 24 11:59:36 2023 +0100
+
+ Add missing license header.
+
+[33mcommit efe77f4260b560243ba300472064dfcb5c347a72[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 24 09:52:33 2023 +0100
+
+ Add "docs" extra deps and remove "all" from "tests".
+
+[33mcommit a038bd9fdf199bab560e46e244a0be0423f30742[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 24 09:39:07 2023 +0100
+
+ Move 'gen_docs.py' to the 'scripts' folder.
+
+[33mcommit 4f7f292ece6cd5b1d4c8ea5e1b130d5866ccdf87[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 23 12:57:08 2023 +0100
+
+ Refactor developers guide into multiple files.
+
+[33mcommit 2a13933e27ebe4050715a158df6033be464a6b38[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 23 12:18:26 2023 +0100
+
+ Add docstrings style-guide to the developers guide.
+
+[33mcommit 44b688d709cd8a1002e2725b93c79df58eec9914[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 23 10:08:43 2023 +0100
+
+ Fix "Raises" and "Warns" sections' formatting.
+
+[33mcommit 1faf2a2433d3b5790bbd9301ab30cd35a571ce91[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 23 09:50:12 2023 +0100
+
+ Fix multi-references rendering.
+
+[33mcommit 131eb4e512053e7ef7fcd5de597fced7c0d79e23[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 22 18:13:57 2023 +0100
+
+ Add LaTeX rendering to the mkdocs config.
+
+[33mcommit de829563723257e2bc690b8774fcf1ce513712fc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 22 18:13:29 2023 +0100
+
+ Improve docstrings' formatting based on mkdocs rendering.
+
+[33mcommit fa55e88cb5c3e2a455c5c9473aa9f065c53b6d86[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 22 14:51:44 2023 +0100
+
+ Revise formatting of module docstrings for mkdocs parsing.
+
+[33mcommit c1eecc64221dfe44ff5e068cedc031121970d3c4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 22 14:48:37 2023 +0100
+
+ Update the API reference docs generation policy.
+
+[33mcommit 35b5fa4660d92181d3c361d238332d02807e1a06[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 16 11:38:19 2023 +0100
+
+ Add information on building the docs.
+
+[33mcommit 9723f8e2c1299adea076e7c93e84a4e848042708[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 15 13:21:23 2023 +0100
+
+ Add website- and API-Reference-generation tools.
+
+ * Write a `gen_docs.py` script that automatically updates the
+ `docs/index.md` file based on the `README.md` one, and more
+ importantly creates the `docs/api-reference/` markdown files
+ by parsing through the declearn source code.
+ * Set up a `mkdocs.yml` file to render the documentation (both
+ static and dynamically-generated) into a website.
+ * Add `SUMMARY.md` files and enhance `index.md` ones to benefit
+ from the mkdocs-literate-nav plugin when building the website.
+
+[33mcommit b980c9330a9fb93201407ccebfb7de561bdd5863[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 15 13:20:50 2023 +0100
+
+ Add "docs" folder and distribute README content within it.
+
+ * Split the increasingly-long README content into multiple files
+ and structure them under the new "docs" folder.
+ * Improve the formatting of some of the files, notably based on
+ their rendering by the mkdocs website-generation tool.
+ * Write a placeholder file for the API reference.
+
+[33mcommit 066407647c1f1a126e48b17121160a95efbc5a8c[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Mar 15 09:30:00 2023 +0100
+
+ Checking process is alive before closing in multiprocessing
+
+[33mcommit 7d08a3c506cecf97ed990d9b74e23d9bebaa47eb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Mar 14 18:03:20 2023 +0100
+
+ Fix 'localhost' specification in tests for the CI.
+
+[33mcommit 9951b76ecd7febfac2f4dc7e225b7f2a702c95e0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 13 17:28:07 2023 +0100
+
+ Re-implement unit tests for communication routines.
+
+[33mcommit ccaf21daced87ccb702fcad454d5343f4a75d98e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 6 16:39:03 2023 +0100
+
+ Fix 'WebsocketServer.stop'.
+
+[33mcommit 957d40d7652902dc9e6c3595d06cd94dd44cd90d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 6 16:38:35 2023 +0100
+
+ Minor docstring formatting fix.
+
+[33mcommit e71cabba642688c1ff82cb7090a0353a7d025c57[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Mar 3 18:23:48 2023 +0100
+
+ Implement generic unit tests for 'NetworkServer'.
+
+ Note: these tests somewhat fall into the issue of being redundant
+ with the souce code, especially as they do not really test that
+ communications really work.
+
+ The next effort will therefore be to try and rewrite client-server
+ routines unit tests, after which the present tests will be skimmed.
+
+[33mcommit 5fe5ed7ca8af9c86c726878df9337feef79f95cc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Mar 14 16:44:15 2023 +0100
+
+ Add a '--cpu-only' mode for unit tests.
+
+[33mcommit 45def4ed8707d80e50f412397749a781c258069e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Mar 14 16:43:39 2023 +0100
+
+ Ensure multiprocessed functional tests run on CPU.
+
+[33mcommit fce0bb75d2da36b87234ed55c470e5f1997ca395[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Mar 14 16:43:14 2023 +0100
+
+ Enable running `test_regression.py` with missing frameworks.
+
+[33mcommit 9ce1ef7d827323767c56d05104533e067af25e3b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 13 17:48:35 2023 +0100
+
+ Add missing license header to 'test_regression.py'.
+
+[33mcommit 9229751096aeb9acbfba777ee0dc0df42766c4d6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Mar 13 15:33:04 2023 +0100
+
+ Fix some docstring typos and errors.
+
+[33mcommit e2b855e4838b1435e79113bbb9a29f7f49fa2000[m
+Merge: fb86f06 9dd5fcc
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Mar 13 14:10:47 2023 +0000
+
+ Merge branch 'improve-multiprocessing' into 'develop'
+
+ Improve `run_as_processes` to capture exceptions and outputs.
+
+ See merge request magnet/declearn/declearn2!37
+
+[33mcommit 9dd5fcc4712b72631129cf4577a4350b57c58f74[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Mar 13 14:10:47 2023 +0000
+
+ Improve `run_as_processes` to capture exceptions and outputs.
+
+ * Add exceptions and outputs catching to multiprocessed routines.
+ * Change the output signature of `run_as_processes` to return a bool
+ flag indicating success, and a list of routine-wise output value or
+ RuntimeError (that may either result from an actual failure or from
+ the process having been interrupted due to another one's failure).
+ * Add `auto_stop` parameter to enable disabling the default automated
+ interruption of processes once any of them has failed.
+
+[33mcommit fb86f06661014bf2f90a910d2b767f1b58929835[m
+Merge: 35ec6c7 afaee4f
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Mar 8 15:01:57 2023 +0000
+
+ Merge branch 'jax-dataset' into 'develop'
+
+ Adding data_type to DataSpecs
+
+ See merge request magnet/declearn/declearn2!36
+
+[33mcommit afaee4f028bb1c742660d92165ce2e83bb46ea0e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 8 14:47:54 2023 +0100
+
+ Mitigate and document 'DataSpecs' API changes.
+
+[33mcommit 5864510d39aa11954cbe99feb0b894efe9c57f2f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 8 12:48:39 2023 +0100
+
+ Polish up data-info fields' revision.
+
+ Data-info fields:
+ * Rename "single_input_shape" to "features_shape".
+ * Move deprecation warnings to `combine` classmethod as the data-
+ info fields are never instantiated.
+ * Make these deprecation warnings more verbose.
+ * On the side, remove some redundant-with-parent type hints.
+
+ Models:
+ * Have "features_shape" verification happen for both regression
+ and classification variants of `SklearnSGDModel`.
+ * Remove requirement for "n_samples" in `TensorflowModel`.
+
+[33mcommit 0f4eced5d78b8da09c0db2db8e8cf4ced1a68f0c[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Mar 7 12:15:10 2023 +0100
+
+ Applying corrections after MR review:
+ * Re-introducing deprecated `DataField` subclasses with a warning
+ * Other minor changes, to `sklearn` and `tf` models, and `DataTypeField`
+
+[33mcommit 83204307d99a0e72631572f835e5267a0e7c79a9[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 6 18:19:06 2023 +0100
+
+ Replacing `n_features` by `single_input_shape` and propagating changes in relevant files
+
+[33mcommit a4da6fdbb9a3251b3df4fd9406d377fcedde5ed7[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Mar 6 16:24:48 2023 +0100
+
+ Correcting tf code (in part) to not use data_shape, for reference purposes
+
+[33mcommit 38cf844fa8c0642c2304d13d3b0db12558d91edd[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 2 14:26:44 2023 +0100
+
+ V0 of adding data_type to DataSpecs
+
+[33mcommit 35ec6c7275b061dce44dbeabce05c23f1d4a509c[m
+Merge: 1d6f499 c6447a5
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Thu Mar 2 14:47:04 2023 +0000
+
+ Merge branch 'modularize-heart-example' into 'develop'
+
+ Modularize "heart-uci" example for its real-life deployment.
+
+ See merge request magnet/declearn/declearn2!34
+
+[33mcommit c6447a5d176a29fe6fbe61b37e56581629f1d713[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 2 14:56:19 2023 +0100
+
+ Minor typos
+
+[33mcommit 1d6f4992211a346b0549c20dda420a8449ffff91[m[33m ([m[1;33mtag: v2.1.0[m[33m)[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Mar 2 09:46:36 2023 +0100
+
+ Bump version to 2.1.0.
+
+[33mcommit be743fe9466a98d99102af35517c02df0b55526b[m
+Merge: 2babbf1 300a1a8
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Mar 2 08:36:07 2023 +0000
+
+ Merge branch 'toy-regression' into 'develop'
+
+ Functional test of regression
+
+ See merge request magnet/declearn/declearn2!28
+
+[33mcommit 300a1a81f881750bf50937a79e298c6186e0b923[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Feb 21 10:51:43 2023 +0100
+
+ Document 'penalty="none"' option for 'SklearnSGDModel'.
+
+[33mcommit 478d8c4e5bb0e7008daf57183d152ce70f0e6d2b[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Feb 13 10:32:13 2023 +0100
+
+ Add a toy regression functional test, checking convergence on simple data
+
+[33mcommit 2babbf18b13e894b87ce6217966c338811b104d1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 1 16:15:26 2023 +0100
+
+ Fix 'TensorflowModel.apply_updates'.
+
+[33mcommit 62683d6336dc7f9af767e30addb85e775fd73800[m
+Merge: fec1fa0 0585e4c
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Mar 1 14:08:39 2023 +0000
+
+ Merge branch 'branching' into 'develop'
+
+ Revise git branching strategy.
+
+ See merge request magnet/declearn/declearn2!35
+
+[33mcommit 0585e4ca3c63468cb4a03b457a23a0c7dbd6f9c0[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Mar 1 14:08:39 2023 +0000
+
+ Revise git branching strategy.
+
+ * Keep "develop" as principal (~main) branch where to push new features,
+ hence acting as a nightly stable version.
+ * Drop "main" branch from the strategy as its name was misleading.
+ * Introduce "rX.Y" release branches, that need protection and CI/CD.
+
+[33mcommit fec1fa06192ce64d173ea3394bb0800451cb05db[m
+Merge: 89f940c ec4f3e1
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Mar 1 13:36:06 2023 +0000
+
+ Merge branch 'indexed-slices' into 'develop'
+
+ Enhance support for 'tf.IndexedSlices' in 'TensorflowVector'.
+
+ Closes #17
+
+ See merge request magnet/declearn/declearn2!33
+
+[33mcommit ec4f3e17b80d6df6945e8ea3990fda11b468b6b0[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Mar 1 13:36:06 2023 +0000
+
+ Enhance support for 'tf.IndexedSlices' in 'TensorflowVector'.
+
+ * Implement a public util to wrap tensorflow operations in order to preserve
+ `tf.IndexedSlices` structures and run appropriate computations with them.
+ * Deploy the former wrapper to cover all usual operations in the backend of
+ `TensorflowVector`.
+ * Refactor the use of the device-placement-handling wrapper in the backend of
+ `TensorflowVector` to reduce runtime overheads and factor it with the new
+ indexed-slices-handling wrapper.
+
+[33mcommit 89f940cedd00b353155de9d31fa57d8f76eaf85f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Mar 1 14:17:28 2023 +0100
+
+ Add a seed to 'test_rsquared.py' to avoid random failures due to numerical precision.
+
+[33mcommit 13459d5981739e70b8cb6d0a578fac6f6988d3d8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 4 16:25:34 2023 +0100
+
+ Deploy `make_importable` to the Heart-UCI example scripts.
+
+[33mcommit b7f2f1f7e9637a82f4da27a6faa4d05b09aafd5e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 4 16:24:56 2023 +0100
+
+ Implement the `make_importable` test-util context manager.
+
+[33mcommit 7a37aa52cb5b78d34800551b29cc5d8381487885[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Nov 25 16:28:07 2022 +0100
+
+ Refactor argparse code from the heart-uci example.
+
+[33mcommit d73f2331de5fd590aeca9c40c4db18381aa6f820[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 24 17:16:19 2022 +0100
+
+ Modularize "heart-uci" example for its real-life deployment.
+
+ * Add optional parameters to the Python functions and the argparser
+ so as to enable running the UCI-Heart-Dataset examples over an
+ actual network.
+ * Make the documentation more verbose about these options, and the
+ difference between localhost-based simulations and real-network
+ use.
+ * Rationale: having a minimal example to run is a valuable feature
+ when deploying declearn to a new machines~network setting, so as
+ to quickly assess (and/or debug) the framework's usability.
+
+[33mcommit 71debe68a72e4a067e8d09d4080c04e96ff7c6f6[m
+Merge: 10d8a40 622d241
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Feb 28 15:48:00 2023 +0000
+
+ Merge branch 'scaffold-update' into 'develop'
+
+ Change 'ScaffoldClient.collect_aux_var' behavior on unused module.
+
+ See merge request magnet/declearn/declearn2!31
+
+[33mcommit 10d8a40db4fa67967d388fcce5f7a1d249c073a1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 27 17:11:41 2023 +0100
+
+ Optimize 'TensorflowModel.apply_updates'.
+
+[33mcommit 915bd4280766009b0b383c321674e77292529378[m
+Merge: 3c82f5c 76ed2a4
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Thu Feb 23 17:13:14 2023 +0000
+
+ Merge branch 'hacky-optim' into 'develop'
+
+ Implement framework-specific OptiModule subclasses.
+
+ See merge request magnet/declearn/declearn2!15
+
+[33mcommit 76ed2a406f4aeefc46fc2bdee5fb1303c33e0b93[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 23 16:43:27 2023 +0100
+
+ Improve 'declearn.model' docstring and add some missing '__all__'.
+
+[33mcommit 3d63f26d89446b3008f7038cef0f06738c2625b8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 23 16:39:01 2023 +0100
+
+ Implement 'TensorflowOptiModule' optimizer plug-in.
+
+[33mcommit 16511fc8bddbf99ba62c7423c85e532157c83c72[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 23 16:35:54 2023 +0100
+
+ Implement 'TorchOptiModule' optimizer plug-in.
+
+[33mcommit 206302d06e855426311abee14f2454b961141038[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 23 16:34:56 2023 +0100
+
+ Refactor 'TestOptiModule' to enable subclassing.
+
+[33mcommit 861447a25dd0edcab55208d278d70a62d4ff0c69[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 23 16:27:19 2023 +0100
+
+ Fix optimizer plug-ins' framework-equivalence test.
+
+[33mcommit 2a61cd47923c381e4c85bcc42222cfd866e526bf[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 23 15:25:17 2023 +0100
+
+ Fix the (un)packing of scalar tensors in 'Vector' subclasses.
+
+[33mcommit a6fc24859a565ec5c3ab39b4fcbb64d7b559193c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 23 16:19:09 2023 +0100
+
+ Use new assertion utils in existing unit tests.
+
+[33mcommit 0420531a9e50b28d1d6214e4da53ef60ada29cd7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 23 15:24:09 2023 +0100
+
+ Implement assertion functions that complexify 'a == b'.
+
+ The implemented functions are a convenience to reduce boilerplate
+ code in unit tests, and improve the existing JSON-serializability
+ check.
+ They notably support comparing numpy arrays (optionally tolerating
+ small absolute discrepancies), and ignoring tuple-list type change
+ that typically result from JSON-serialization.
+
+[33mcommit 1fbecfacfed035d64d6fb316a9c8a62fc9553094[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 23 15:23:49 2023 +0100
+
+ Add a generic functional test for `OptiModule.set_state`.
+
+ * Add `test_set_state_results` to verify that resetting a modules' state
+ enables running the same computation twice. This should enable detecting
+ cases when information is missing from the returned state dict.
+
+ * Note that at the moment noise-addition modules are ignored. We could look
+ into a way to access and restore RNG states (when CSPRNG is not used).
+
+[33mcommit 74d6808f29b56987fc93149f59bee33c38ff27d7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 12 17:48:42 2023 +0100
+
+ Adjust TensorflowVector square and square root computations.
+
+ For some reason, `tf.square(x)` and `tf.pow(x, 2)` can result in
+ distinct values for small-enough values of x. The same is true of
+ `tf.sqrt(x)` and `tf.pow(x, 0.5)`.
+
+ This commit ensures the `tf.sqrt` and `tf.square` functions are
+ used in these two edge cases for the `vector ** p` syntax. This
+ choice is motivated by the fact that these functions are used
+ in most official tensorflow / keras code, and seem to be closer
+ to their torch counterparts (which are equivalent to `vec ** p`)
+ than `tf.pow` - based on a few local tests.
+
+[33mcommit 3c82f5ce719668a64d0af4ccce6fd7b9c6c0187a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Feb 22 18:35:53 2023 +0100
+
+ Make network communication dependencies optional.
+
+ This removal was supposed to be part of a past commit
+ (cd1d0cb137b74c6baec397c703f4d1c8830427c4) but had in
+ fact not been properly finished.
+
+ With this commit, install declearn no longer results
+ in the automatic installation of gRPC and websockets,
+ which are properly relegated to being optional, user-
+ triggered-installation, dependencies.
+
+[33mcommit dba20501715e4498b9f58dbb0f02d67d4407586e[m
+Merge: ae0c079 28865db
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Feb 22 14:53:56 2023 +0000
+
+ Merge branch 'handle-frozen-weights' into 'develop'
+
+ Enable skipping frozen weights when using `Model.get_weights` and `Model.set_weights`
+
+ Closes #15
+
+ See merge request magnet/declearn/declearn2!29
+
+[33mcommit 28865db8e5b7a6209e8951ea5982a3301d8ca772[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Feb 22 14:53:55 2023 +0000
+
+ Add 'trainable' argument to 'Model.get_weights' and 'Model.set_weights'.
+
+ This commit squashes the following modifications:
+
+ * Add `trainable` argument to `Model.get_weights`:
+ - Add `trainable: bool = False` argument to `Model.get_weights`,
+ enabling accessing only trainable weights of a neural network
+ (or any model that supports using frozen weights).
+ - Use the former to fix support for models with frozen weights by
+ the `Optimizer`'s weight-decay and loss-regularization plug-ins.
+ - Add unit tests for `get_weights(trainable=True)` to the Torch
+ and Tensorflow Model tests.
+
+ * Add `trainable` argument to `Model.set_weights`:
+ - Add `trainable: bool = False` argument to `Model.set_weights`, enabling
+ to update model weights with the exclusion of frozen ones.
+ - Improve the documentation and exception-raising of `Model.set_weights`.
+ Notably refactor some code into a private util used by tensorflow and
+ torch models' backend code.
+ - Add some related tests to the tensorflow and torch Model test suites.
+ Note that there is redundancy that could be tackled later, as part of
+ an effort to improve the `Model` test suite template.
+
+ * Restrict round-wise weights sharing to trainable weights:
+ - Update `FederatedClient`, `FederatedServer` and `TrainingManager`
+ so that only the trainable model weights' values and updates are
+ communicated during training and evaluation rounds.
+ - This reduces communication costs when some weights are frozen.
+ - This also avoids including non-trainable weights in server-side
+ optimization computations (which should have resulted in zero at
+ any rate, properly leaving these weights unaltered -- but may not
+ have done so in the rare case when weight decay or regularization
+ was used on the server side).
+
+ * Add 'MLP-tune' test case for 'TorchModel' and 'TensorflowModel':
+ - This case is the same as the base MLP, but with a frozen first layer.
+ - This case would fail if the `get_weights(trainable=False)` was badly
+ implemented (or used), since the gradients-and-weights comparison in
+ the `Model.apply_updates` test would raise an error.
+
+ * Improve unit test for 'Optimizer.compute_updates_from_gradients':
+ - Ensure the pipelined plug-ins were called in proper order and
+ with proper input values. In other words, test the formula.
+ - Ensure the model weights are only accessed once, and with the
+ proper restriction to trainable weights.
+
+[33mcommit ae0c0796531b468b101a82bcfa10371316d51019[m
+Merge: 73f4d1e f370dad
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Feb 22 14:27:37 2023 +0000
+
+ Merge branch 'gpu-support' into 'develop'
+
+ Improve GPU support for TensorFlow and Torch
+
+ Closes #11
+
+ See merge request magnet/declearn/declearn2!24
+
+[33mcommit f370dadcf03d709574170ea1c2eef9286398b07a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Feb 22 14:23:14 2023 +0100
+
+ Fix 'Vector' backend to enable 'numpy + tensor' operations.
+
+[33mcommit 5c1dc54af469202de6917846aa3e9398632a2f6f[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Tue Feb 21 17:13:29 2023 +0100
+
+ Improve the documentation of GPU support and of 'Vector' subclasses.
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+[33mcommit cce29e1ac2facef777d457f1e9339f1adea520b7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 16 13:17:13 2023 +0100
+
+ Document GPU support in the README.
+
+[33mcommit 996b22a636da5de015c645bce2b6c1038727beb3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 16 12:34:44 2023 +0100
+
+ Enhance unit tests to take device-placement policy into account.
+
+ Core changes to the global test suite:
+ * Add some device-placement verifications to the generic Model test
+ suite. This should be refactored into more unitary tests as part
+ of a distinct effort to revise and improve this test suite.
+ * For TensorFlow and Torch, parametrize the whole test suite to run
+ either on CPU and GPU, and run it once per available device type.
+ * Ensure Vector and OptiModule unit tests run on CPU.
+
+ Changes to some 'Model' tests to run on GPU:
+ * `test_compute_batch_gradients_np`: allow for small numerical
+ discrepancies that may result from running the test on GPU.
+ * `test_apply_updates`: correct the test, that had not been
+ updated since `Model.get_weights` stopped to systematically
+ use `NumpyVector` as return type.
+
+ Changes to 'TorchModel' tests to run on GPU:
+ * Override `test_serialization` due to torch-serialization relying
+ on pickle. Replace pickles' comparison with a more shallow (but
+ less susceptible to fail for unknown reasons) test that ensures
+ a reloaded model shares the same structure of modules.
+
+[33mcommit 098c26e6b6f2a8fc1c26c0658be8e84c5b5f50af[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 16 11:04:29 2023 +0100
+
+ Abstract GPU-related methods into the Model API.
+
+[33mcommit c8a3dbb64e0fa75db51e4a50b6e53edbdbfb0b42[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 10 17:09:36 2023 +0100
+
+ Add GPU support for TensorFlow.
+
+ * Add `declearn.model.tensorflow.utils` submodule. Move existing `build_keras_loss`
+ function into it, and implement the following new utils:
+ - move_layer_to_device:
+ Create a copy of an input keras layer placed on a given device.
+ - preserve_tensor_device:
+ Wrap a tensor-processing function to have it run on its inputs' device.
+ - select_device:
+ Select a backing device to use based on inputs and availability.
+ * Add device-management to `TensorflowModel`:
+ - Copy the input model to (or deserialize reloaded one onto) a device selected
+ based on the global device-placement policy.
+ - Add `with tf.context(device)` instructions in public methods that require it.
+ - Add the `TorchModel.update_device_policy` method to enable changing the backing
+ device for computations (following what was done for `TorchModel`).
+ * Add device-management to `TensorflowVector`:
+ - Ensure all operations applied to the wrapped tensors occur on the device backing
+ these tensors (which may vary from tensor to tensor).
+ - Overall, ensure the specification as to tensors' device-placement are the same
+ as those of the current `TorchVector`, and document them.
+
+[33mcommit c768efed918c70ad1fd8980ecdb78dbf4c7e420d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Feb 9 14:19:45 2023 +0100
+
+ Add GPU support for torch.
+
+ * Add `declearn.model.torch.utils` submodule, that currently provides with:
+ - AutoDeviceModule:
+ Wrapper for a `torch.nn.Module` automating device-management.
+ - select_device:
+ Select a backing device to use based on inputs and availability.
+ * Add device-management to `TorchModel`:
+ - Wrap the model and loss function modules using `AutoDeviceModule`.
+ - Add `tensor.to(device)` instructions in places that require it.
+ - Consult the global `DevicePolicy` as part of `TorchModel.__init__`
+ so as to automatically place the wrapped model and computations.
+ - Add the `TorchModel.update_device_policy` method to enable changing
+ the device policy to an input one or to the global one (that may have
+ changed since the model was first instantiated).
+ * Add device-management to `TorchVector`:
+ - Ensure `TorchVector.__eq__` transcends device-placement.
+ - Ensure combining with a `NumpyVector` works regardless of device-placement.
+ - Implement a device-placement rules when combining two `TorchVector`: move
+ tensors from the right-hand vector to the device backing the left-hand one.
+ - Document the former rule, as well as the current limitation that device-
+ placement information is lost through serialization, resulting in reloaded
+ vectors being placed entirely on CPU.
+ - Ensure unpacking a `TorchVector` from a serialized dump results in placing
+ it on a device that abides by the global `DevicePolicy`.
+
+ Note on the consequence to support GPU-backed training and evaluation:
+ * A `TorchModel` placed on GPU will handle moving its input data there, so that
+ output gradients are backed on GPU (but output numpy predictions are moved to
+ CPU).
+ * The processing of gradients into updates will be kept on GPU as long as there
+ is no `output = operation(cpu_vector, gpu_gradients)` operation in that flow.
+ The latter could for example happen as part of an optimizer plug-in that uses
+ deserialized state (or auxiliary variable) vectors, in the rare case when the
+ global device policy is not in line with that of the model.
+ * Input updates to a `TorchModel` may be placed on any device, and will preserve
+ its initial placement (whether via `apply_updates` or `set_weights`).
+
+[33mcommit dd029d78e4bbb763c5c6710ee1b545365ba08904[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 10 12:12:55 2023 +0100
+
+ Add device-placement policy utils.
+
+[33mcommit 73f4d1eb361f56da01b445bf10faee5322f11583[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 20 14:34:19 2023 +0100
+
+ Add SAN capabilities to 'generate_ssl_certificates'.
+
+ * Add the `alt_ips` and `alt_dns` optional parameters to the
+ `declearn.test_utils.generate_ssl_certificates` function,
+ that enable configuring the created server certificate to
+ cover alternative IP addresses or DNS names, using Subject
+ Alternative Name (SAN) extensions.
+ * Refactor the entire function to ease the readability and
+ maintainance of the backend code.
+ * Improve the function's documentation.
+
+[33mcommit de0bbbc5cbf095b029ea570a07c817069114c2d1[m
+Merge: 30f1230 cdd004f
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 20 09:49:21 2023 +0100
+
+ Merge branch 'main' into develop
+
+[33mcommit 30f1230f2fee1ff364ddb9bffb8708039aeb5418[m
+Merge: 430be0f 9a9ffe7
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Feb 17 12:08:30 2023 +0000
+
+ Merge branch 'r2' into 'develop'
+
+ Implement the 'RSquared' metric
+
+ See merge request magnet/declearn/declearn2!30
+
+[33mcommit 9a9ffe7d769a4f17ba95231cb7d78c0f9a45d3e1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 17 12:15:10 2023 +0100
+
+ Deprecate 'Metrics.normalize_weights'.
+
+[33mcommit 647a59b07e317d3d9c900749a5c0cc26710106a7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 17 12:01:11 2023 +0100
+
+ Fix formatting of Metric test files.
+
+[33mcommit 062f55840faae0f1290d57642a9c03df2885ad0b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 17 11:57:12 2023 +0100
+
+ Finalize RSquared Metric implementation and unit tests.
+
+ * Rename R2 to RSquared out of compliance with PEP-8.
+ * Improve states' naming and documentation to improve readability.
+ * Rename backend files out of clarity.
+ * Add RSquared to the `declearn.metrics` module docstring.
+ * Refactor copy/pasted code from MeanMetric to RSquared.
+ * Improve unit tests:
+ - use regression-like data and actual sample weights
+ - use scikit-learn to compute the reference score (test our maths)
+ - remove redundant or useless code to set up the test suite
+
+[33mcommit ab1320f1f20073a1475fa6422e500116826f23ee[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Feb 13 15:17:07 2023 +0100
+
+ First implementation of R2 Metric.
+
+ * This squashed commit comprises the result from an iterative effort
+ to implement the R-squared evaluation metric for regression tasks.
+ * It comprises a working implementation of the metric, as well as
+ associated unit tests.
+ * It does not pass linter tests.
+
+[33mcommit 622d24169613f9af9a98f9e3b15412e0e0b95368[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 17 12:24:04 2023 +0100
+
+ Change 'ScaffoldClient.collect_aux_var' behavior on unused module.
+
+ * Until now, calling the `collect_aux_var` method of a `ScaffoldClient`
+ module that had not processed any gradients raised a RuntimeError.
+ * This commit changes that behavior for something less stringent:
+ warn a RuntimeWarning that such a call is unexpected, but return
+ a conventional, zero-valued scalar state.
+ * Conversely, upon receiving such states, `ScaffoldServer` ignores the
+ client (i.e. it is not part of the global state's update calculation).
+ * The rationale behind this change is that it may happen that a client
+ does not perform any optimization step, e.g. due to their DP budget
+ having been saturated. In that case, we do not want the FL process
+ to crash, but rather the client to be ignored as part of the update
+ process.
+
+[33mcommit cdd004f7e053f4887feb8dda67e1a731c00fb835[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Feb 14 10:17:24 2023 +0000
+
+ Fix the main docstring of 'BinaryRocAUC'.
+
+[33mcommit 430be0f9cdde30b065523eeb5b7aea6933b45ce1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 13 15:34:50 2023 +0100
+
+ Add 'develop' branch to our git branching strategy.
+
+[33mcommit efbd18d0dc86530f22455d0fab25edad96af3fe1[m[33m ([m[1;33mtag: v2.0.1[m[33m)[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 13 10:30:41 2023 +0100
+
+ Update version to 2.0.1
+
+[33mcommit 18238f188200a52aebd570f1b9c73518933082e8[m
+Merge: e8afbe1 1e388de
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Feb 13 09:15:10 2023 +0000
+
+ Merge branch 'hotfix-warn-frozen-weights' into 'main'
+
+ Warn about frozen weights use in `TensorflowModel` and `TorchModel`.
+
+ See merge request magnet/declearn/declearn2!27
+
+[33mcommit 1e388de1aeae7a0c730858e946b911f9d36cffbf[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 10 19:01:14 2023 +0100
+
+ Warn about limited support for frozen neural network weights.
+
+[33mcommit e8afbe1b6d8471d8efb33ad94da89318579a95bb[m
+Merge: 7155419 261ab37
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Feb 10 17:51:20 2023 +0000
+
+ Merge branch 'hotfix-tf-gradient-names' into 'main'
+
+ Fix `TensorflowModel` gradients' labeling.
+
+ Closes #14
+
+ See merge request magnet/declearn/declearn2!26
+
+[33mcommit 261ab37917e4f164e3f1d1f3ca996ece3f953c71[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 10 18:17:58 2023 +0100
+
+ Add gradients-labeling check to the Model test suite.
+
+[33mcommit c55a3265bfcc4bd5a86a965d18bd342e99ea6b98[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 10 18:09:44 2023 +0100
+
+ Fix the labeling of output gradients in `TensorflowModel`.
+
+[33mcommit 715541977806752e6b14570fbc54000acf524fda[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Feb 8 10:18:14 2023 +0100
+
+ Add license header to 'test_vector.py'.
+
+[33mcommit 108463839e8abac16427991a7d59a35a93edfbe9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Feb 7 12:07:52 2023 +0100
+
+ Garbage-collect unused 'type: ignore' comments.
+
+[33mcommit cd8e394fe8b1dfa4c6965bae86e7a2a268d636d0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Feb 7 11:40:59 2023 +0100
+
+ Update mypy to version 1.0 and deploy Self type-hint.
+
+[33mcommit 203dce04c946fc8c082b7478995a8fd92ed1f3d9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Feb 7 11:09:22 2023 +0100
+
+ Correct some docstrings to comply with PEP 257.
+
+[33mcommit 111339961f23183112d3bbc95a37d1794ddef808[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Feb 7 10:11:43 2023 +0100
+
+ Update installation instructions.
+
+[33mcommit 22f6b996cec4cab51fee1a0783d2965530cdafdb[m[33m ([m[1;33mtag: v2.0.0[m[33m)[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 17:26:05 2023 +0100
+
+ Fix pypi classifiers.
+
+[33mcommit 0237f15137a6cf8ab0b5721c73ea1cf1ed14b227[m
+Merge: ddc0402 8b08248
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Feb 6 16:06:13 2023 +0000
+
+ Merge branch 'release-todo' into 'main'
+
+ Release version 2.0
+
+ See merge request magnet/declearn/declearn2!23
+
+[33mcommit 8b0824831546d2e1cf94f9d7051b838b1a8526b2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 16:46:49 2023 +0100
+
+ Update pyproject.toml metadata.
+
+[33mcommit 90a6f0ee7aa64488d25f0959e5d2491f677c68f8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 16:45:26 2023 +0100
+
+ Add a missing 'type: ignore' comment.
+
+[33mcommit 57419237345991b3d5a68cab6601db192f315dc8[m
+Merge: 92e82db 9ee99bc
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Feb 6 15:40:43 2023 +0000
+
+ Merge branch 'test-vector' into 'release-todo'
+
+ Adding unit tests for vectors
+
+ See merge request magnet/declearn/declearn2!22
+
+[33mcommit 9ee99bc8c4db1292f5d80d72040df6fc07dd26fe[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Feb 6 15:40:43 2023 +0000
+
+ Adding unit tests for vectors
+
+[33mcommit 92e82dbff6c02542ce39281bb8f8322f3d34098c[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Feb 6 15:23:40 2023 +0100
+
+ Reviewing doc
+
+[33mcommit 0077dde9db8aa515c3631c2d19743eeb13b3541d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 15:07:51 2023 +0100
+
+ Add AUTHORS file.
+
+[33mcommit d91a42565d18ea2467f916331bb194a0a7bd584e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 15:01:37 2023 +0100
+
+ Update package version and classifiers.
+
+[33mcommit b72856ac795927f95e57784f7cab4d837e3c99a6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 14:58:17 2023 +0100
+
+ Enhance some `Vector` docstrings.
+
+[33mcommit 95c4191e3f8e43164e94c020dbd76682ea36c4aa[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 12:05:32 2023 +0100
+
+ Add Apache-2.0 license file, headers and information.
+
+[33mcommit 6360878f27a504818ca1a5429ce5ec4a42a1fb05[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 11:12:10 2023 +0100
+
+ Modularize tests' execution based on frameworks' availability.
+
+[33mcommit d846802c5a809b6b0bf2cd69c6f085ae8999e403[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 10:00:11 2023 +0100
+
+ Drop support for Python 3.7.
+
+ Support for Python 3.7 had been envisioned and partially documented,
+ but had not been tested for a while. As it turns out, some of the
+ `typing` functionalities the generic TOML-parsing system relies upon
+ were only introduced in Python 3.8. Since rewriting this code would
+ be tedious without these utils, and since official support for py37
+ ends in less than six months, it was decided not to support it for
+ declearn.
+
+[33mcommit 5c8c649917aad05eb4dd3b4cf80d53d8a26c5d41[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Feb 6 09:51:42 2023 +0100
+
+ Add some missing 'type: ignore' comments.
+
+[33mcommit b3b75a7aa970c15a383ccdf306676e23f492db16[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 3 17:38:39 2023 +0100
+
+ Update black to version 23.
+
+[33mcommit b752723ceadf984335d7fdd8506af0155defc691[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Feb 3 15:14:08 2023 +0100
+
+ Fix some errors in `NumpyVector.sum` and `TorchVector.maximum`.
+
+[33mcommit a8fa357c19248119cb91f62c475a59f738f4c864[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 3 15:12:35 2023 +0100
+
+ Collect some unused statements.
+
+[33mcommit 06f0092b4eeebcc0d4a7f8f9f49aeba22d8c7156[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Feb 3 15:02:22 2023 +0100
+
+ Remove deprecated `Model.compute_loss` method.
+
+[33mcommit 464e1c416c91a3aadea2ab48ac365ac6d83f9c93[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jan 16 17:41:57 2023 +0100
+
+ Fix some docstring and code typos.
+
+ * Fix FLRunConfig handling of `register=<int>` inputs.
+ * Fix docstring typo in `OptiModule`.
+ * Fix typos in server-side code sample in README file.
+
+[33mcommit 7bd717fd3040f021b5963e9a432f3839f65a857a[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Feb 2 16:12:15 2023 +0100
+
+ Remove superflous `return NotImplemented` and `return None` in abstract methods.
+
+ Note: kept the `raise NotImplementedError` statements.
+
+[33mcommit 2db25bbe33fc9249d70b647da025be96f282f7f4[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Feb 2 15:54:44 2023 +0100
+
+ Rename some unexposed functions to make them private.
+
+ Reviewed all 58 `__all__`, found no error, renamed some methods
+ as private since they were not used outside of the file.
+
+[33mcommit 87cc9d7a9972b5514ea5bffc4495ff6f785999d7[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Feb 2 15:34:36 2023 +0100
+
+ Add `super().__init_subclass__()` calls.
+
+ This commit implements minor changes to stick to the python doc
+ (https://docs.python.org/3/reference/datamodel.html?highlight=__init_subclass__#object.__init_subclass__).
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+[33mcommit aee79fc11ae62af7dc936d203af5bc7d9015b015[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Thu Feb 2 13:49:12 2023 +0100
+
+ Add NoReturn type-hints
+
+ NoReturn type is only meant for functions that never return normally, so
+ quite a rare occurence (https://peps.python.org/pep-0484/#the-noreturn-type).
+
+ Manually scanned the 250 raise statements. Only changed two `_raise_...` methods
+ in `main/utils/data_info.py`.
+ Other places were it could be relevant:
+ * `message_pb2_grpc.py` > big ' DO NOT EDIT" in there so did not touch
+ * `api/_vector.py` > abstract methods raising a `NotImplementedError`,
+ the current convention seems to be using the expected implementation
+ type, so I did not change
+
+[33mcommit 21f9b0d96efcfa2f0286bc96db56aeb61bf93d80[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Wed Feb 1 17:29:25 2023 +0100
+
+ Add ClassVar type-hints.
+
+ Details :
+ * In `/optimizer`, `/metrics`, and `/aggregator` :`name` and `aux_name`
+ * In `/communications` : `protocol`
+ * In `/dataset` : `_type_key`
+ * In `/data_info` : `field`, `type`, and `doc`
+
+ Excluding all classes decorated with `@dataclasses.dataclass`,
+ see https://stackoverflow.com/a/52099457
+
+[33mcommit ddc04028c8424d4e45ebb2950b0bd521c3ea6bac[m
+Merge: 5f23f93 4cf8b54
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Thu Feb 2 14:26:19 2023 +0000
+
+ Merge branch 'revise-checkpointer' into 'main'
+
+ Revise `Checkpointer`.
+
+ See merge request magnet/declearn/declearn2!21
+
+[33mcommit 4cf8b5496b9437c575b31eee5b85a04995a3f590[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Thu Feb 2 14:26:18 2023 +0000
+
+ Revise `Checkpointer`.
+
+ * Rewrite the `declearn.main.utils.Checkpointer` util.
+ - Implement Model, Optimizer and Metric-output results saving and loading methods.
+ - Add a `max_history` parameter to garbage-collect past files.
+ - Revise the labeling of output files to ease collection and re-use.
+ * Revise the use of `Checkpointer` in `FederatedClient` and `FederatedServer`.
+ - Have both entities checkpoint at the end of evaluation rounds.
+ - For now, clients save the global model (not local one), local optimizer and metrics.
+ - For now, server saves the global model (not local ones), server optimizer and all
+ metrics: aggregated ones and client-wise ones (in separate files).
+ - Move lowest-loss-yielding weights tracking and restoring to the private backend
+ of `FederatedServer`, and ensure they are still saved and sent to clients at the
+ end of the federated learning process.
+ * Add unit tests for the `Checkpointer` class.
+
+[33mcommit 5f23f9351af634c7a0a5493afc7081bb0f02ad8d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Feb 1 14:50:57 2023 +0100
+
+ Minor pylint-based syntax fix.
+
+[33mcommit b302cd2f4403490f00c72a30bcb57b16f1d2b664[m
+Merge: bd45c0a 8b40601
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Feb 1 13:40:36 2023 +0000
+
+ Merge branch 'metrics' into 'main'
+
+ Implement Metric API.
+
+ See merge request magnet/declearn/declearn2!20
+
+[33mcommit bd45c0ad95a4bf5a9b937ab35a5d3f01d20ca2df[m
+Merge: 7c0dca2 d933ac7
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Feb 1 13:36:43 2023 +0000
+
+ Merge branch 'stateful-optimizer' into 'main'
+
+ Revise `Optimizer` and add state-access methods.
+
+ See merge request magnet/declearn/declearn2!19
+
+[33mcommit 7c0dca2daa4f5fc1fc9390a39379e753388a0649[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 31 12:33:10 2023 +0100
+
+ Fix side effect in `SklearnSGDModel.set_weights`.
+
+[33mcommit 8b406014de19394bb59b4e2fcecba153b92bfc97[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Jan 27 16:30:37 2023 +0100
+
+ Implement `Metric.normalize_weights` util.
+
+[33mcommit d933ac706883cce4d1e9ea7ab423c4d8f0fee5fe[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 12:13:40 2023 +0100
+
+ Patch `test_train_manager.py` to avoid failure on slower machines.
+
+[33mcommit 8a528ae09953fc7f13bd41ed3a0a22be63a04a9e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 12:13:40 2023 +0100
+
+ Patch `test_train_manager.py` to avoid failure on slower machines.
+
+[33mcommit 2bca6bdbe71d8c123eb0ab940075479a94e76b28[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 12:40:36 2023 +0100
+
+ Update the RMSProp-implementation example.
+
+[33mcommit 35cde15994c358d6535c0b20620bf554cbbfc8b3[m
+Merge: c9be886 1bd8a5f
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 12:02:14 2023 +0100
+
+ Merge branch 'main' into metrics
+
+[33mcommit c9be88624636360c3644cb84f8a03d5e9587c827[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 11:48:08 2023 +0100
+
+ Add metrics computation to the Heart UCI example.
+
+[33mcommit 8ac4aec99b95a9ec3989e0c1251892304ecbb89d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 11:43:33 2023 +0100
+
+ Update README to document the Metric API.
+
+[33mcommit 358872bfea68a23ec10e2b3258f08200b3e2b08c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 11:21:09 2023 +0100
+
+ Add `share_metrics` bool parameter to `FederatedClient`.
+
+ * Enable clients to prevent sharing evaluation metrics (and their
+ intermediate states for federative aggregation) with the server.
+ * Enable server to handle clients' replies when metrics are lacking.
+
+[33mcommit 90ad4c574e6abfb39201c49fabc986e0e950306e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 11:21:05 2023 +0100
+
+ Update unit tests for `TrainingManager`.
+
+[33mcommit 41cf12874986fd2979ef94a9e72da72d5a34a318[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 10:51:29 2023 +0100
+
+ Minor docstring and code-order revisions to `FederatedServer`.
+
+[33mcommit 892a68b26bdbb0d073783e2e607847089de4b765[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 10:46:10 2023 +0100
+
+ Enable server-driven computation of federative evaluation metrics.
+
+ * This commit adds an optional `metrics` parameter to `FederatedServer`
+ that enables specifying evaluation metrics that are to be computed by
+ each and every client during evaluation rounds.
+ * The resulting metric states are then shared by clients with the
+ server, that procedes to their exact aggregation thanks to the
+ `Metric.agg_states` method.
+ * Note that this commit does not (yet) implement:
+ - proper saving of the computed metrics, which for now are merely
+ and clumsily logged to the console
+ - privacy safeguards enabling clients to control which metrics are
+ computed and shared with the server (as intermediate metric states
+ might in some cases leak information, e.g. as to the distribution
+ of target labels or values)
+
+[33mcommit e2018f8e5dd8503ba4338c3578695e7dbc198b27[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 10:43:31 2023 +0100
+
+ Refactor some code into `MetricSet.from_specs`.
+
+[33mcommit 5df95e7592b08aa628d5b4462895092b794e7fa0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 27 09:59:50 2023 +0100
+
+ Patch `test_optimizer.py` to silence pylint.
+
+[33mcommit 7e059884d9fc1b3c97be86c405aad3103e3a3229[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 26 17:55:48 2023 +0100
+
+ Clean up `OptiModule` code by removing unused statements.
+
+[33mcommit 9c3267f5e692dbc2cade02a901cbafb93200b898[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 26 17:48:22 2023 +0100
+
+ Remove out-of-use OptiModule (de)serialization methods.
+
+[33mcommit 13f6304dd799e0c6ae7ccb90c7c93a11718576db[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 26 17:25:24 2023 +0100
+
+ Add `Optimizer.start_round` method to trigger `Regularizer.on_round_start`.
+
+[33mcommit 47ae6cc4fdf488edc0c0ef6fac609004badf7616[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 26 17:09:15 2023 +0100
+
+ Implement unit tests for `Optimizer`.
+
+[33mcommit ab7e2559a3b26842e6b12611d6c1b9d78c3b42ec[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 26 14:47:34 2023 +0100
+
+ Add `assert_json_serializable_dict` to `declearn.test_utils`.
+
+[33mcommit ed342b82f1130651ee7a339cd47020f0712136cb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 26 14:41:40 2023 +0100
+
+ Deploy `access_types_mapping` to unit test code.
+
+[33mcommit 162f80525f6f3221af36a546cc1f6d8c31227cbe[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 26 10:39:24 2023 +0100
+
+ Revise the format of `Optimizer`'s config and states.
+
+ * The udpate to the config format enables including multiple instances
+ of a plug-in in the Optimizer pipeline (which is unlikely to happen,
+ but was a limitation neither documented nor motivated).
+ * It also removes some redundant code in `Optimizer.from_config`.
+ * The update to the states format covers the same concern, but goes
+ along an extension of the verifications and error-handling of the
+ `set_state` method.
+
+[33mcommit a6d7d03d05f8dd163ab2bb3b8308836b2161e3c5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 25 18:18:14 2023 +0100
+
+ Fix YogiModule algorithm.
+
+[33mcommit 4d6c65942e2027e63235ad02c945300b52c2318f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 25 18:22:56 2023 +0100
+
+ Improve some `OptiModule` docstrings.
+
+ * Improve the docstrings of `get_config`, `from_config` and `run`.
+ * Remove overriding docstring from subclasses, to ease code readability
+ and maintenance.
+
+[33mcommit 44f834343546e032e93ad0a02633f94ccb5252c2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 25 18:28:13 2023 +0100
+
+ Add unit tests for `OptiModule.(get|set)_state` methods.
+
+[33mcommit 993d8eef8bc8a54d4865b1ad2dc826e9928d4064[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Jan 16 18:11:20 2023 +0100
+
+ Add `Optimizer.get_state` and `set_state` methods to the API.
+
+ * Add `get_state` and `set_state` to the `OptiModule` API.
+ * Add the same methods to `Optimizer`, that collects and propagates
+ information from and to its plug-in modules.
+ * These methods, which are unused for now, enable saving the state
+ of the optimizer between rounds, which can be useful to repeat
+ or resume training from a given point when running experiments.
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+ Authorship:
+ * Nathan Bigaud wrote the first version of this commit.
+ * Paul Andrey cleaned it up and applied fixes for missing variables.
+
+[33mcommit 2c424b35fcc5e1430db5e42f3669ac375ea27497[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 25 15:53:43 2023 +0100
+
+ Handle division-by-zero errors in metrics and add associated tests.
+
+[33mcommit 522bfd3344201a58da18605d4cdc86f076cffc3b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 25 11:36:54 2023 +0100
+
+ Add F1-Score to classification metrics.
+
+[33mcommit 1bd8a5ff9cbf8b229214025a5bd88600da200fc4[m
+Merge: feac22d f880486
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Jan 25 08:06:15 2023 +0000
+
+ Merge branch 'revise-network-api' into 'main'
+
+ Revise some front-end aspects of network endpoints
+
+ See merge request magnet/declearn/declearn2!18
+
+[33mcommit 374324182408352130194de977cb15c7484d82dc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 24 15:34:57 2023 +0100
+
+ Patch Metric tests to comply with linters.
+
+[33mcommit 9f68f9cda43b289d99eb5dcecb2968c5408d6563[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 24 09:52:52 2023 +0100
+
+ Minimally deploy the Metrics API to `TrainingManager`.
+
+ * This commit adds a `metrics` attribute (and instantiation parameter)
+ to `TrainingManager`, that enables defining evaluation metrics, in
+ addition to the model's loss.
+ * The backend of `TrainingManager.evaluation_round` is also updated to
+ make use of this `MetricSet` and of the newly-introduced inference
+ methods of `Model`.
+ * At the moment, the `EvaluationReply` message does not support adding
+ computed metrics beyond the loss.
+ * Similarly, the "main" centralized FL process does not yet provide
+ with the possibility to specify metrics to be computed.
+ * Hence, this commit:
+ - adds the potentiality and some core API modifications to compute
+ evaluation metrics modularly
+ - revises the way the model's loss is computed, without altering its
+ logic nor the way the results are handled in the FL process
+
+[33mcommit bef9ef8be0cb43d72a7dc13e5806adaa63785cac[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 24 12:19:51 2023 +0100
+
+ Update Model API for better articulation with the Metric API.
+
+ * Add `Model.compute_batch_predictions`, interfacing the wrapped model
+ in inference mode.
+ - This method outputs numpy arrays so as to enable passing its output
+ to a `Metric` or `MetricSet` instance.
+ - It may also be overridden by end-users to alter the flow between
+ inputs and outputs, e.g. for self-supervised learning models.
+ * Add `Model.loss_function` to finalize sample-wise loss computation.
+ * Abstract `Model.compute_loss` into the base class using the previous
+ two methods. Deprecate it, in favor of out-of-Model orchestration.
+ * Expand Model unit tests to cover these new functions.
+
+[33mcommit 6fc008960a5a46936dbda41c2e7f419e02d2eeac[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 18 13:41:42 2023 +0100
+
+ Implement `MetricSet` wrapper for `Metric` instances.
+
+ * Implement `MetricSet`, a wrapper for an ensemble of `Metric` instances.
+ * Design `MetricSet` to mimic the `Metric` API, similar to the way how an
+ `Optimizer` wraps a pipeline of `OptiModule` plug-ins.
+ * Design `MetricSet` to be instantiable from a human-editable and JSON-
+ serializable config, similar to the way an `Optimizer` can be.
+ * Expose a `MetricInputType` type-alias for the variety of input formats
+ enabling to specify the individual wrapped metrics, to ease source code
+ readability.
+ * Implement unit tests for `MetricSet`.
+
+[33mcommit f20c14f6591c7340da1bc1deed016e78d09d4262[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 24 11:05:47 2023 +0100
+
+ Implement standard classification and regression metrics.
+
+ * Implement accuracy, precision and recall metrics for classification,
+ covering both the binary and multi-class classification cases.
+ * Implement binary roc-auc metric, with modular precision on thresholds
+ used for fpr/tpr estimation, including privacy-oriented restrictions.
+ * Implement abstract `MeanMetric` template for averaged scalars.
+ * Implement `MeanAbsoluteError` and `MeanSquaredError` metrics.
+ * Implement associated unit and functional tests, including a shared
+ test suite that may be re-used for future `Metric` subclasses.
+
+[33mcommit 4ebb627852f3981ca4fafc27576c4b8f496268dc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 24 10:59:03 2023 +0100
+
+ Implement Metric API.
+
+ * Add `declearn.metrics` submodule.
+ * Implement `Metric` abstract base class, that defines an API for
+ the iterative and federative computation of evaluation metrics.
+ * This commit is the result from an iterative process:
+ - Implement a first draft of the Metric API, with some subclasses.
+ - Marginally revise the API based on subclasses' implementation
+ and test-driven debugging.
+ - Harmonize the Metric API with the type-registration and name-based
+ specification practices used in the Regularizer and OptiModule APIs.
+
+[33mcommit feac22d72b62788a53af5352ee418995f3c69f40[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jan 23 09:34:15 2023 +0100
+
+ Add 'tomli' dependency for Python <3.11.
+
+[33mcommit f880486463a771f1efee3e69dd2dd080c7167b1b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jan 16 09:50:35 2023 +0100
+
+ Add TOML-parsing to network endpoints' build-config dataclasses.
+
+ * Use dataclass-generation utils to replace the hand-written
+ `NetworkClientConfig` / `NetworkServerConfig` dataclasses.
+ * Use inheritance from `TomlConfig` to add TOML-parsing.
+ * Update `FederatedClient` and `FederatedServer` to use the
+ revised classes' features (including TOML parsing).
+
+[33mcommit cd1d0cb137b74c6baec397c703f4d1c8830427c4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 11 14:34:11 2023 +0100
+
+ Made gRPC and websockets network communication backends optional.
+
+[33mcommit 54b3f2ee7d883bbc48f43dbbfde6bbfa4231ae07[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 11 14:24:49 2023 +0100
+
+ Add `declearn.communication.list_available_protocols`.
+
+[33mcommit b0f880a6eaedc1fb27820990b71e6793681b71a5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 11 13:53:59 2023 +0100
+
+ Add `access_types_mapping` util to access registered type mappings.
+
+[33mcommit f64067e079f84eaeaf514fb038de653129dfe59a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 11 13:11:47 2023 +0100
+
+ Rename `Client` and `Server` to `NetworkClient` and `NetworkServer`.
+
+ On the side, slightly update the API of these classes, in a similar
+ fashion as what was recently done for `OptiModule` and `Regularizer`:
+ * Automate type-registration of child classes.
+ * Add a `protocol` str class attribute, used by the automated type
+ registration.
+
+[33mcommit c841bfcb66587d1f9adbbd4b105d9eb7da9f7f2f[m
+Merge: d2ef427 d4cd449
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Jan 17 08:35:14 2023 +0000
+
+ Merge branch 'revise-strategy-api' into 'main'
+
+ Revise strategy api
+
+ See merge request magnet/declearn/declearn2!16
+
+[33mcommit d4cd449fa240124032a2b6884b63d18f00e2249d[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Jan 16 19:25:02 2023 +0100
+
+ minor name chnages and typo
+
+[33mcommit d2ef427e3e8518559818e05af71d5f8e965147de[m
+Merge: 8824105 29043ba
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Jan 16 18:20:25 2023 +0000
+
+ Merge branch 'hotfix-examples' into 'main'
+
+ Hotfix examples
+
+ See merge request magnet/declearn/declearn2!17
+
+[33mcommit 34c6e4ec573c358f1b2bbe6c5dded0171c5b5d0e[m
+Merge: 973167f 8824105
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jan 16 11:33:27 2023 +0100
+
+ Merge branch 'main' into revise-strategy-api
+
+[33mcommit 8824105071787c74f38cc8ed3aba51714a98e92f[m
+Merge: 168781e 37cfaf4
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Jan 16 09:13:06 2023 +0000
+
+ Merge branch 'dp-sgd' into 'main'
+
+ Add local DP to declearn
+
+ See merge request magnet/declearn/declearn2!11
+
+[33mcommit 29043ba4d34d5bc663f393c593e42d4f84392504[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 12 15:59:49 2023 +0100
+
+ Fix an argparse bug in the Heart-UCI server script.
+
+[33mcommit 9571ad560cde8fc76ebc37661c9ef5b1c6c82734[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 12 15:59:09 2023 +0100
+
+ Update RMSprop-addition example.
+
+[33mcommit 37cfaf4a4aeaca1f4be0ba3644a31a63e244109d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jan 11 11:08:47 2023 +0100
+
+ Fix mypy error due to untyped opacus imports.
+
+ Note: depending on the opacus (and, seemingly, python) version used,
+ opacus's type hints are available to mypy or not. This commit hence
+ takes a conservative approach at silencing errors that mypy might in
+ some contexts not raise.
+
+[33mcommit 168781e2cbe1cfe0b68913818aaa187271312a47[m
+Merge: b796921 b8c969c
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Jan 9 16:14:58 2023 +0000
+
+ Merge branch 'refactor-main' into 'main'
+
+ Refactor `FederatedServer` and `FederatedClient` code.
+
+ See merge request magnet/declearn/declearn2!14
+
+[33mcommit fc50779e2f71462320a2a360bff141e00aecd42a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jan 9 10:00:48 2023 +0100
+
+ Increment version number and add Nathan to authors.
+
+[33mcommit 973167fd04e09aaf86eb9df40b4b7e499a0d75b4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Jan 9 09:54:52 2023 +0100
+
+ Enable specifying `FLOptimConfig` optimizers as single floats (lrate).
+
+[33mcommit 1a6e937e57b786d92f7a95ac04725cb56c0aa179[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 6 18:10:55 2023 +0100
+
+ Remove `declearn.strategy` in favor of using `FLOptimConfig`.
+
+[33mcommit e4d28bee6a9526a1c21a3c233e91045bfd42e9cd[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 6 18:05:59 2023 +0100
+
+ Implement `FLOptimConfig` as an alternative to the Strategy API.
+
+[33mcommit f4d19036113ec32ee67954e6ed94eada15f07d52[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jan 6 12:03:31 2023 +0100
+
+ Move Aggregator API and subclasses to the new `aggregator` module.
+
+[33mcommit d27fcaf9de61276d5803dd8c060056b439ed8dda[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 5 17:04:49 2023 +0100
+
+ Marginally improve TomlConfig docs and type-hints parsing.
+
+[33mcommit 81e530ed354f2cbafee9aad98e306098b38babb0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jan 5 15:10:20 2023 +0100
+
+ Implement unit tests for `declearn.main.privacy.DPTrainingManager`.
+
+[33mcommit 8b5e8cd4c92f2122fc096edce85c563164ea63b2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 3 15:49:06 2023 +0100
+
+ Implement DP-budget-based training interruption in `DPTrainingManager`.
+
+ * `DPTrainingManager`'s (private) training step method now implements
+ the verification of the DP budget that would be spent as a result
+ of the step, and effectively prevents the step from being taken if
+ needed.
+ * To achieve the former, `TrainingManager._train_under_constraints`
+ now catches any built-in `StopIteration` errors raised when calling
+ `_run_traing_step`, and interprets it as a signal to exit the loop.
+ * The latter behaviour is now documented at the base class level.
+ * Note: in case the train step is authorized but fails, the accountant's
+ history will be wrong - but so may be the optimizer's plug-ins' states
+ etc. hence this is not tackled as of now.
+
+[33mcommit d5170c9d16fc6c2fc45ba61fa466cc45f4e91fe6[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Dec 12 15:04:50 2022 +0100
+
+ Add documentation on local DP, including warnings and limits.
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+[33mcommit 46981363dff0967dc3d13b072603348fef22381f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 3 11:39:27 2023 +0100
+
+ Document local DP setup in the README.
+
+[33mcommit 6a39d72a0ff5321e1d36afab1c1304b54873255f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Jan 3 10:14:30 2023 +0100
+
+ Modularize available frameworks' listing and use in optimizer tests.
+
+ * Replace the hard-coded `Frameworks` tuple with a `list_available_frameworks`
+ function that checks third-party dependencies' availability.
+ * Avoid importing all frameworks when importing `declearn.test_utils`, which
+ resulted in slow start-up (and unrequired warnings) of examples code.
+ * Adjust the optimizer tests based on available frameworks.
+
+ Notes:
+ * Additional work is required to deploy this function to other tests.
+ * It may be useful to move `list_available_frameworks` to `declearn.model`
+ (or `declearn.utils`?).
+
+[33mcommit 26bebdc1ba45400b85f50473d8969813656040a9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Dec 5 14:55:50 2022 +0100
+
+ Fix `TorchVector` compatibility with `NumpyVector`.
+
+[33mcommit 3e1a2ae1b9539a731ba84fbc8589c19b17da5064[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Dec 23 17:27:13 2022 +0100
+
+ Enable DP-SGD setup in `FederatedClient` and `FederatedServer`.
+
+ * Add an optional secondary initialization phase to set DP-SGD up
+ based on the presence of a "privacy" field in the `FLRunConfig`.
+ * Have the server package and send information required to set up
+ the local DP.
+ * Have the client replace its `TrainingManager` with the DP-SGD
+ subclass `DPTrainingManager` and call its `make_private` method.
+
+ Co-authored-by: BIGAUD Nathan <nathan.bigaud@inria.fr>
+
+ Authors:
+ * Nathan Bigaud first implemented the setup and use of DP-SGD by
+ subclassing `FederatedClient` and `FederatedServer`.
+ * Paul Andrey made use of the revisions inspired by Nathan's work
+ to come up with this final implementation.
+
+[33mcommit 4738265c3a897b7c2e202fc9c78366d0b6488a41[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Dec 23 17:14:07 2022 +0100
+
+ Implement `DPTrainingManager` to enable DP-SGD use.
+
+ * Implement `DPTrainingManager`, a subclass of `TrainingManager` that adds
+ DP-SGD tools and enables setting them up through a `make_private` method.
+ * Expose it under a dedicated `declearn.main.privacy` submodule so as to
+ lazy-import 'opacus'. Also implement informative warnings when opacus is
+ missing.
+ * NOTE: some "# REVISE" comments should be tackled before this commit hits
+ the main branch.
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+ Authors:
+ * Nathan Bigaud wrote a first version of DP-SGD-enabling code, using the
+ opacus dependency and its `make_private` naming logic.
+ * Paul Andrey re-implemented from this code using the newly-introduced
+ `TrainingManager` and `sclip` parameter to `Optimizer.run_train_step`.
+
+[33mcommit c97a6510359cca626dfdc062defba4c66e659151[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Dec 23 16:58:33 2022 +0100
+
+ Add a PrivacyRequest message and marginally modify InitRequest.
+
+ * PrivacyRequest is a new message designed to wrap up hyper-parameters
+ required to set up local differential privacy (using DP-SGD).
+ * InitRequest has been provided with a 'dpsgd' bool field (default=False)
+ that will serve in the future to indicate that initialization should
+ be pursued by a local-DP-setup phase (using a secondary PrivacyRequest).
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+ Authors:
+ * Nathan Bigaud first implemented PrivacyRequest.
+ * Paul Andrey polished it and made the change to InitRequest.
+
+[33mcommit 3cb81145cb24eb177d3772ee0044949b088ac941[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Dec 23 16:55:36 2022 +0100
+
+ Add `PrivacyConfig` dataclass and make it an optional section of `FLRunConfig`.
+
+ Co-authored-by: BIGAUD Nathan <nathan.bigaud@inria.fr>
+
+ Authors:
+ * Nathan Bigaud wrote the first implementation of PrivacyConfig.
+ * Paul Andrey polished it and articulated it with FLRunConfig.
+
+[33mcommit 05c9c6195319a29dc41672b8e8dcddb219171213[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Dec 23 11:32:15 2022 +0100
+
+ Minor type-hinting fix in Model tests code.
+
+[33mcommit 3b2cfdbd72b32094b608323f65ad8b860594aa76[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Dec 23 11:28:57 2022 +0100
+
+ Refactor some optimizer unit test code, moving some to `declearn.test_utils`.
+
+ * Move up the listing of frameworks and the mock-gradient-vectors-generating
+ util class to a new private submodule of `declearn.test_utils`.
+ * Add a shared framework-providing fixture to the optimizer tests' module.
+ * Harmonize relative imports' code (preserving the `sys.path` trick).
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+[33mcommit 7faf34ea8fc926e1d8a02da82a9f53dcbb22674e[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Dec 23 11:26:56 2022 +0100
+
+ Implement functional tests for `NoiseModule` subclasses.
+
+[33mcommit a50b810f72d04701b55ac8d19ab2c854d01ff039[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Dec 23 11:22:28 2022 +0100
+
+ Implement abstract and gaussian noise-addition optimizer modules.
+
+ * Implement `GaussianNoiseModule`, to enable adding gaussian noise
+ to step-wise batch-averaged gradients as part of local DP-SGD or
+ to clients-averaged updates as part of central DP-SGD.
+ * Implement an abstract `NoiseModule` layer, paving the way for
+ other noise-addition mechanisms.
+
+ This commit squashes some incremental work by multiple authors:
+ * Nathan Bigaud implemented `GaussianNoiseModule` after doing the
+ scientific and technical review of requirements, notably as to
+ the use of a cryptographically-safe RNG.
+ * Paul Andrey abstracted part of the code into the `NoiseModule`
+ class, fixed some pending todos, and adjusted unit tests.
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+[33mcommit b978cfaa15dd1a4cc7e3b186e99071cd37ddbece[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Fri Nov 25 19:10:17 2022 +0100
+
+ Add Poisson sampling option to `Dataset.generate_batches`.
+
+ * Add `poisson` argument (bool, default=False) to `Dataset.generate_batches`.
+ * Implement Poisson sampling for `InMemoryDataset`.
+ * Add `poisson` parameter to the `TrainRequest` dataclass, enabling instructing
+ clients to use Poisson sampling rather than default batching in a FL process.
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+[33mcommit b8c969c1a9202c1894110b92bb398f18d7e8b7d0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Dec 22 16:13:59 2022 +0100
+
+ Patch backend code for Python <3.9 compatibility.
+
+[33mcommit 04360a84c079867c5b911d399d0185f343329c21[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Dec 22 13:59:32 2022 +0100
+
+ Implement unit tests for `declearn.main.utils.TrainingManager`.
+
+[33mcommit a84f40aa0981bc414a0c24541a1eab7350912d35[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Dec 22 13:58:59 2022 +0100
+
+ Fix bug: infinite training round loop when `epochs=None`.
+
+[33mcommit 53cd2d775444710b8f86c001dcfe71f207e614c5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Dec 13 16:38:13 2022 +0100
+
+ Refactor `FederatedClient`, splitting out training and evaluation loops.
+
+ * Move local training and evaluation routine under a new class,
+ `TrainingManager`, that retains objects expected to have been
+ created based on server-issued instructions.
+ * Have `FederatedClient` use the former class, using an attribute
+ set during the initialization phase.
+ * As a result, this commit splits the client-side code between:
+ - training and evaluation orchestration, as synchronous methods
+ that may be used and tested locally (even outside of FL, even
+ though the signatures currently retain messaging features).
+ - initialization and message-based-processing orchestration,
+ that sets up the former and makes use of it based on server
+ instructions, and transmits replies over the network.
+
+[33mcommit bb133eb45b01b1af0b2f8877514a3db808fb4369[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Dec 21 16:03:54 2022 +0100
+
+ Clean up some FederatedServer backend code.
+
+[33mcommit 171b691b592cee1dea250332bc4e1ee2a1c063cc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Dec 21 15:35:57 2022 +0100
+
+ Refactor `FederatedServer` by introducing `FLRunConfig`.
+
+ * Add `declearn.main.config` submodule:
+ - Move dataclasses (`TrainingConfig`, `EvaluateConfig`
+ and `RegisterConfig`) to this new submodule.
+ - Implement `FLRunConfig`, that binds together all arguments and
+ hyper-parameters used by the `FederatedServer.run` method, and
+ provides with the possibility to write them up as a TOML config
+ file (by inheriting `declearn.utils.TomlConfig`).
+ * Deploy `FLRunConfig` to `FederatedServer`:
+ - Replace the argument to `run` and `async_run` by `FLRunConfig`,
+ adding the possibility to either provide (legacy-style) kwargs
+ or the path to a TOML file to the `run` one.
+ - Replace the argument to `initialize` by the full `FLRunConfig`,
+ to facilitate overriding by subclasses that may need to send or
+ use information on the process in addition to conducting client
+ registration.
+ - Preserve the use of targetted fields of hyper-parameters in the
+ `training_round` and `evaluation_round` methods.
+
+[33mcommit e85d1184104a5541698d14801c4e6183a4e2358e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Dec 12 16:36:03 2022 +0100
+
+ Simplify server-side hyperparameters' collection into messages.
+
+[33mcommit f814d6c720dced91c99b3f970a3454e0928f9424[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Dec 12 11:18:10 2022 +0100
+
+ Use `dataclass_from_init` to implement `EarlyStopConfig`.
+
+[33mcommit ede01beb8a4b7e86ba57005abd6234ffdc1a5814[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Dec 9 17:26:00 2022 +0100
+
+ Implement dataclass-generation tools.
+
+[33mcommit c1e0bf9d13f69280413d47bb39ff754f1434da89[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Dec 21 15:01:39 2022 +0100
+
+ Implement `TomlConfig` util to define TOML-parsable configuration containers.
+
+[33mcommit b7969219953d453a3baf032acea51698f247489a[m
+Merge: d153178 c0340d9
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Dec 21 09:28:19 2022 +0000
+
+ Merge branch 'clip-gradients' into 'main'
+
+ Add sample-wise gradients clipping to the Model API
+
+ See merge request magnet/declearn/declearn2!13
+
+[33mcommit c0340d910f92622d37d8b758f83e28c94b73a44d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Dec 21 09:44:17 2022 +0100
+
+ Implement (batch-averaged) gradient clipping module.
+
+[33mcommit 8a5602de4c1e8f8f1088505955ea616cfb3777bb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Dec 21 09:42:56 2022 +0100
+
+ Add and implement to the API.
+
+[33mcommit 0db226a6070a1c017a1af7ddfb64cd95d3a923e6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Dec 20 13:03:54 2022 +0100
+
+ Add `compute_updates_from_gradients` method to the `Optimizer` API.
+
+[33mcommit c9224fff362e025d0a62fb8751dfd593892b549a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Dec 20 12:52:38 2022 +0100
+
+ Add `sclip` sample-wise clipping parameter to `Optimizer.run_train_step`.
+
+[33mcommit c51ae84d3f85601d38a18e0ebbaa2eb2de84506d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 5 12:30:57 2022 +0200
+
+ Implement unit tests for sample-wise gradients clipping.
+
+[33mcommit 35bdadd501f52692c188b417833ff491b59f7aa7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 5 12:30:39 2022 +0200
+
+ Add sample-wise gradients clipping to the Model API.
+
+[33mcommit 44accf61c16bfec1c46ec2ea3f937fe760435ee9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 5 12:30:26 2022 +0200
+
+ Add sample-wise gradients clipping to `TorchModel`.
+
+ Note: to efficiently compute (and clip) sample-wise gradients
+ in Torch, the additional dependency `functorch` is used. This
+ poses a risk as to the maintainance of `declearn[torch]` due
+ to `functorch` still being presented as experimental by Meta;
+ however it is the best available solution to limit overheads,
+ and will probably become an important component of the torch
+ ecosystem in the future.
+
+[33mcommit 8528f8e200aedf896d2dcb8148deb5155b8b87b6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 5 12:27:29 2022 +0200
+
+ Add sample-wise gradients clipping to `TensorflowModel`.
+
+[33mcommit 56bb8da128856fe1d28e616f6b121a478d91d16a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 5 12:27:16 2022 +0200
+
+ Add sample-wise gradients clipping to `SklearnSGDModel`.
+
+[33mcommit d1531784155a2128265784f1215f8c74d682689d[m
+Merge: fa2adab dd4e4f6
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Dec 14 08:54:14 2022 +0000
+
+ Merge branch 'gradient-masked-averaging' into 'main'
+
+ Gradient Masked Averaging
+
+ See merge request magnet/declearn/declearn2!12
+
+[33mcommit fa2adab3d86b96ca658ca636948450d090d40b46[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Dec 8 08:36:25 2022 +0100
+
+ Reformat extra dependencies' specifications.
+
+ The previous version specified some recursive dependencies, to avoid
+ writing redundant dependency specifications. This is convenient, and
+ supported (as of versions 19.0 to 22.3) by pip. However, this is no
+ longer working with tox v4.0.2 (as opposed to v3.27), that was just
+ released.
+
+ As a consequence, and since self-dependency is not part of PEP 631,
+ this commit removes it, and introduces some (documented) redundancy
+ in extra dependencies - which as of now make for a couple of lines
+ only.
+
+[33mcommit f42f7ed7b634b0ae94bc82dc90752415d76c26e0[m
+Merge: 01ead20 f2a2116
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Dec 9 08:53:30 2022 +0000
+
+ Merge branch '12-chunk-large-messages-to-avoid-size-limitations' into 'main'
+
+ Chunk large messages to avoid size limitations
+
+ Closes #12
+
+ See merge request magnet/declearn/declearn2!10
+
+[33mcommit f2a21161061885e66558349417eb7ee9d276b51b[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Dec 9 08:53:30 2022 +0000
+
+ Chunk large messages to avoid size limitations
+
+ This commit implements message-chunking in cases when large (usually, >1MiB)
+ messages are being exchanged via one of the current network communication
+ protocols (WebSockets or gRPC). The code additions are framework-dependent,
+ but are implemented in coherence, following these common specs:
+
+ * When a (string-serialized) message reaches a size that exceeds the
+ communication protocol's (default) message-size limitation, it is sent
+ as a stream of ordered chunks, obtained by merely splitting the string
+ into fixed-size chunks.
+ * This may be used either on the client or server side, and should be
+ invisible to the user (including developers): chunking is automated as
+ part of the backend of public message-sending methods.
+ * When an unregistered client attempts to send a chunked message to the
+ server, its message should be refused, without collection nor parsing
+ of the actual message contents.
+
+ The practical implementations are the following:
+ * WebSockets:
+ * When chunking, send a conventional flag as first message; expect an
+ acceptance (or denial) flag; sequentially send chunks; send a closing
+ flag.
+ * When receiving, check for the "stream start" flag; decide whether to
+ accept (based on registration status) and send a flag; receive and
+ assemble chunks until the ending flag is received.
+ * In practice, util functions have been implemented, that are shared by
+ the Client and Server classes (but not made available to the end-users).
+ * gRPC:
+ * Turn `send` into a server streaming RPC, enabling the server to send
+ replies as an ordered sequence of message chunks.
+ * Add the `send_large` bidirectional-streaming RPC, that is similar to
+ `send` but is used when the client performs message-chunking.
+ * Refuse `send_large` requests from unregistered clients (send back an
+ `Error` message).
+ * Automate the choice of `send_large` over `send` in the client backend.
+
+[33mcommit dd4e4f6622355aa8c2f3cadedc58415c7a1d81a4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Dec 8 08:36:25 2022 +0100
+
+ Reformat extra dependencies' specifications.
+
+ The previous version specified some recursive dependencies, to avoid
+ writing redundant dependency specifications. This is convenient, and
+ supported (as of versions 19.0 to 22.3) by pip. However, this is no
+ longer working with tox v4.0.2 (as opposed to v3.27), that was just
+ released.
+
+ As a consequence, and since self-dependency is not part of PEP 631,
+ this commit removes it, and introduces some (documented) redundancy
+ in extra dependencies - which as of now make for a couple of lines
+ only.
+
+[33mcommit af10d54b5395858657d0def1c61385cb2c5bc08f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Dec 7 17:57:11 2022 +0100
+
+ Fix `Vector.__rsub__` method.
+
+[33mcommit 0513abdd246377db96afd6ee8e8f1d2107ee71c4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Dec 7 17:28:46 2022 +0100
+
+ Implement GradientMaskedAveraging aggregator.
+
+[33mcommit 01ead2049a8dffe005b59168d1d5067505fb1ece[m
+Merge: 750df01 8bee8cd
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Dec 2 11:20:17 2022 +0000
+
+ Merge branch 'regularizers' into 'main'
+
+ Implement loss regularizers & revise Optimizer API.
+
+ See merge request magnet/declearn/declearn2!9
+
+[33mcommit 8bee8cd53cb45dcc630c83b3a5aa47b06f962ba5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Dec 1 16:19:26 2022 +0100
+
+ Fix Lasso and Ridge regularizers.
+
+[33mcommit a46dc0ed8c152e914f1b0579fc856c6c23329547[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Nov 22 11:04:40 2022 +0100
+
+ Update CI/CD configuration.
+
+ * Prevent automatic testing on Draft MRs.
+ * Prevent automatic testing on merge commits to main.
+ * Enable manual launch of "fulltest" case on Draft MRs and commits to
+ main.
+
+[33mcommit 57c30306ee2d8d613b445fb5e10fe726a1fab7dd[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 21 17:19:20 2022 +0100
+
+ Increment version to "2.0.0.beta2".
+
+[33mcommit 5e2cda593aa10b63e33eee3872f3676c8001d254[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 21 17:18:52 2022 +0100
+
+ Update Heart-UCI example using new `Optimizer` config syntax.
+
+[33mcommit c8c8a85d8962350c00d0ecca9985425f17dfd490[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 21 17:16:43 2022 +0100
+
+ Document `Regularizer` API in the README file.
+
+[33mcommit 2beb9f2eb1614a8d3ae593b51f2d0c3ddd1ff921[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 21 16:57:33 2022 +0100
+
+ Move `NumpyVector` from `model.api` to `model.sklearn` submodule.
+
+[33mcommit 750df01736bef60e6309958e295b43891a4c4881[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Nov 18 17:48:05 2022 +0100
+
+ Fix `InitRequest.to_string` with tensorflow 2.11.
+
+[33mcommit a63c27b41876c5f2f9bb65ce6b3388ac6cabfa94[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 21 16:41:53 2022 +0100
+
+ Refactor `OptiModule` unit tests and implement `Regularizer` ones.
+
+[33mcommit c88c122bdd1a73cdae7ce476a35deb5e174a9c19[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 21 12:09:42 2022 +0100
+
+ Apply minor docstring and type-hint corrections.
+
+[33mcommit 2ed8e450b1811f66dc2f5c087da20ed52b6ace95[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 21 13:22:15 2022 +0100
+
+ Modularize `regularizers` and `modules` input types for `Optimizer`.
+
+[33mcommit 0887fdd36c49d15ec43040cee5b14bce755d9967[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 21 13:13:56 2022 +0100
+
+ Revise `OptiModule` API.
+
+ * Add `aux_name` class attribute, to distinguish between unique module
+ `name` and auxiliary-variables-tagging name that may be used to pair
+ up a couple of server/client modules (an example being Scaffold).
+ * Revise `Optimizer.process_aux_var` and `Optimizer.collect_aux_var`'s
+ backend in accordance with the addition of `aux_name`.
+ * Revise the (de)serialization of `OptiModule` by adding `from_specs`,
+ similarly to `Regularizers`. Previous (de)serialization methods may
+ be garbage-collected in the near future.
+ * Implement the automatic type-registration of `OptiModule` subclasses
+ using `__init_subclass__`, as was done with `Regularizer`.
+
+[33mcommit 04bf1448b054d190e7e29f14fc83b8c710d01854[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Nov 18 17:48:05 2022 +0100
+
+ Fix `InitRequest.to_string` with tensorflow 2.11.
+
+[33mcommit 53b27196686d21676b42ead5176050166d5b2efe[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Nov 18 15:56:50 2022 +0100
+
+ Use framework-specific Vector type as `Model` weights.
+
+ * Prior to this commit, the API had `Model` subclasses
+ use `NumpyVector` as model weights' container. This
+ felt convenient as few computations were expected to
+ make use of these weights directly.
+ * However, this caused an asymmetry between calls to
+ the server-side and client-side optimizers, and was
+ the cause of implicit (or boilerplate-code-demanding)
+ explicit casting between frameworks as part of the
+ `apply_updates` method.
+ * With this commt, model weights are now wrapped in a
+ framework-dependent container, as are model gradients.
+
+[33mcommit fd1c012d2f937a5717aabbe5757cd07e714b5860[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 17 17:52:34 2022 +0100
+
+ Added loss regularizers to `Optimizer`.
+
+[33mcommit e1479287cd60514d528fbe2d580b716896cbd533[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 17 17:50:00 2022 +0100
+
+ Implement FedProx, Lasso and Ridge loss regularizers.
+
+ On the side, alter `Regularizer` API: move name class attribute definition.
+
+[33mcommit 53dd9eafd8b1469bb1aa1f151deb3843d89ef84b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 17 17:49:30 2022 +0100
+
+ Implement `Regularizer` optimizer plug-ins API.
+
+[33mcommit bde421d1f4aa11dfb3e2d229e2f08d25d278ac67[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 17 17:46:10 2022 +0100
+
+ Rename private `optimizer.modules._base` to `_api` for readability.
+
+[33mcommit 0a1d967f54af1cef402874bbdeabc5d3373b6771[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Nov 17 17:16:40 2022 +0100
+
+ Revise `create_types_registry`.
+
+ * Invert the order of arguments `base` and `name`.
+ * Enable not specifying `name`.
+ * Enable class-decoraotr syntax use of the function.
+ * Propagate these changes throughout `declearn`.
+
+[33mcommit f07d18c209292aebe84ef40e6bd1666fb9e3955a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 16 11:13:33 2022 +0100
+
+ Marginally revise `generate_ssl_certificates`.
+
+[33mcommit f8663243956832fcd285ab9bb5192d4bc8fc62a5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 16 11:12:41 2022 +0100
+
+ Add logging to `WebsocketsClient` failed connection attempts.
+
+[33mcommit b1cd7907ac55d35dca84fa847c908193de3adb10[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 14 17:23:09 2022 +0100
+
+ Fix `TorchModel` gradients' naming.
+
+ Note: unproper naming had a potential nasty side-effect
+ of disabling model update on the server side due to key
+ mismatch between coefficients' names.
+
+[33mcommit cbc02c70e9aa97ad4331b74b6403cb9e6453e2b6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Nov 9 18:58:50 2022 +0100
+
+ Implement and deploy `json_load` and `json_dump` utils.
+
+[33mcommit c3a20cbd80048537a2aadc702986ef5a55cb5646[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Nov 7 12:09:44 2022 +0100
+
+ Correct `Scaffold` client-state update computation.
+
+[33mcommit 41f937ab01962f9c3b907f49c08ef1e636683bb4[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Nov 7 11:01:21 2022 +0000
+
+ Fix Momentum formulation and add Nesterov variant.
+
+[33mcommit 7fd99b75e11ae781c5569c5a85bbc729c97bcfec[m
+Author: BIGAUD Nathan <nathan.bigaud@inria.fr>
+Date: Mon Oct 24 15:23:29 2022 +0000
+
+ Add tutorial on RMSProp implementation.
+
+[33mcommit aeba86495d1f198a493ff656805b371c2083b76c[m
+Merge: 6882651 144c0c3
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Mon Oct 24 15:22:31 2022 +0000
+
+ Merge branch 'use-mypy' into 'main'
+
+ Automate code type-checking using mypy.
+
+ See merge request magnet/declearn/declearn2!6
+
+[33mcommit 144c0c39b9a061dc747864d34d9bc0ebf750a4dc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Oct 24 10:32:23 2022 +0200
+
+ Add some type checks and hints to model subclasses.
+
+[33mcommit 6882651c21a7297be01c6501e9a4b60f1ae099c8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 19 16:08:44 2022 +0200
+
+ Prevent the absence of effort constraints on training and validation.
+
+[33mcommit 537a326928febd8febfd3b8d7f462b9a210dbf26[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Oct 18 08:08:42 2022 +0000
+
+ Update the README file.
+
+[33mcommit 03ec2cb658797018b07f73b11ade48715701c51f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Oct 17 11:50:04 2022 +0200
+
+ Minor formatting fixes.
+
+[33mcommit 2e670270f0f8afb339804fe2ad34a96ed0dd67d1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Oct 17 10:04:55 2022 +0200
+
+ Turn `Vector` operator class attributes into properties.
+
+ The rationale behind this change is that assigning a function
+ to a class attribute results in it being bound as a method as
+ per the Python data model.
+ However, for unclear reasons that may have to do with the way
+ they are built, `operator.[op]`, `numpy.[op]` and `torch.[op]`
+ are not affected by this binding, which is not desirable in
+ the context of `Vector`.
+ Yet, to prevent this issue in future framework-specific classes
+ (which was hackily dealt with for tensorflow), and to improve
+ their documentation, these base operator attributes were turned
+ into properties.
+ Additionally, this commit extends the docstring of `Vector` to
+ provide developers with indications as to what should or may be
+ implemented, extended or overridden by concrete subclasses.
+
+[33mcommit ea984108719b2217cc3d9a6c0cf631e2e2924a3e[m
+Merge: 2e1c1d2 ae5b064
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Oct 14 19:12:00 2022 +0200
+
+ Merge branch 'main' into use-mypy
+
+[33mcommit ae5b064bf7b21959d5d5d041d9024965d8da14f2[m
+Merge: b2a4423 9e9efb3
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Fri Oct 14 16:45:52 2022 +0000
+
+ Merge branch 'heart-demo' into 'main'
+
+ Add examples folder
+
+ See merge request magnet/declearn/declearn2!5
+
+[33mcommit 9e9efb35120bd91c63cd44870c7edb0569d37865[m
+Author: Nathan Bigaud <nathan.bigaud@inria.fr>
+Date: Fri Oct 14 18:31:24 2022 +0200
+
+ Implement UCI heart disease dataset example.
+
+ Co-authored-by: Paul Andrey <paul.andrey@inria.fr>
+
+[33mcommit 06250995407784193c711b8e3055b21166827845[m
+Author: Nathan Bigaud <nathan.bigaud@inria.fr>
+Date: Fri Oct 14 18:28:43 2022 +0200
+
+ Reformat README file.
+
+[33mcommit 4091c5405064c2f7bd6e2356c297c9417eff05b1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Oct 14 18:01:20 2022 +0200
+
+ Implement and deploy `declearn.test_utils.run_as_processes`.
+
+[33mcommit c08c741f13d3e4d7ce813963433c678e391d2136[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Oct 14 18:00:49 2022 +0200
+
+ Add `declearn.test_utils` and move `generate_ssl_certificates` into it.
+
+[33mcommit 2e1c1d2aeeca23adf706ff6f81affa0a9d2df02e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Oct 13 13:59:37 2022 +0200
+
+ Automate code type-checking using mypy.
+
+[33mcommit b2a44235ec0743d240504beaa0bf316b1f0d0ffe[m
+Merge: eeae8c6 6c33d9f
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Oct 12 13:18:06 2022 +0000
+
+ Merge branch 'gardening' into 'main'
+
+ Code Cleaning & Automation
+
+ See merge request magnet/declearn/declearn2!3
+
+[33mcommit 6c33d9f5f80dfc3d4aed373f8be90e3b88b37375[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 12 12:11:33 2022 +0200
+
+ Document `Vector.coefs` attribute in (sub)classes' main docstring.
+
+[33mcommit aebb9c3f98f7d937622a711fd8facec49f54cd4b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 12 12:07:01 2022 +0200
+
+ Revise rng seeding to generate data batches.
+
+[33mcommit 0fcedf3946c27196bfcfb0e20e3544ace803d71c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 12 11:48:43 2022 +0200
+
+ Raise if `InMemoryDataset.generate_batches` returns an empty iterator.
+
+[33mcommit b20071abc7daffee44c01f8d133c975b36cf5c65[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 12 11:40:57 2022 +0200
+
+ Add `dtypes` and `shapes` methods to `Vector` and refactor `__repr__`.
+
+[33mcommit 27b0aa013bfcc7588784a1a1e9854e1bff32a81b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Oct 12 10:18:27 2022 +0200
+
+ Extend README as to black, pylint and conda.
+
+[33mcommit 587c8c23546ddf98c713d7a115e30f8592f48971[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Oct 11 13:02:49 2022 +0200
+
+ Automate code verification using tox and CI/CD.
+
+ * Add `black --check` and `pylint` code verification commands to the `tox.ini` file.
+ * Add the associated tools as extra dependencies for `declearn[tests]`.
+
+[33mcommit 4d793ac2768aed9f59a6d5641d59e6178fbe57a8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Oct 11 14:43:24 2022 +0200
+
+ Add `Vector.compatible_vector_types` property and remove redundant code.
+
+[33mcommit cf2b41dc28184c6a812e71c80ec8d1031e24bf10[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Oct 11 14:22:53 2022 +0200
+
+ Refactor some redundant code in communication Server classes.
+
+[33mcommit 2af0234062ef55d81615b301935269a2e7c07f4f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Oct 11 13:02:00 2022 +0200
+
+ Reformat code files using black.
+
+ * Run [black](https://github.com/psf/black/) to reformat the existing codebase.
+ * Add a limited amount of `# fmt: off` comments to protect some code blocks that
+ are believed to achieve better readability in their previous form.
+ * Add (limited) black configuration parameters to `pyproject.toml` for their use
+ by any contributor to declearn.
+
+[33mcommit 995499c7abc5a901144ea75bd5d83633affd633d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Oct 10 16:51:23 2022 +0200
+
+ Fix `pyproject.toml` installation and drop `setup.cfg`.
+
+ * Fix setuptools instructions as part of `pyproject.toml`.
+ * Drop now-useless `setup.cfg` file.
+ * Add a missing dependency (installed by tensorflow, hence
+ unseen when running unit tests): `protobufs`.
+
+[33mcommit fb6d2109378287a63329e15df957c05ac1e3b7d0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Oct 10 14:18:27 2022 +0200
+
+ Add `logger` argument to communication and orchestrating classes.
+
+ * Remove former `logger` class attributes.
+ * Add `logger` optional argument to `Client`, `Server`, their subclasses and
+ the associated build tools.
+ * Add `logger` optional argument to `FederatedClient` and `FederatedServer`.
+ * The new `logger` argument enables passing an existing logger, or the name
+ of a logger to create or re-use. When left to None, a name is automatically
+ generated, which differs from client to client based on their name.
+ * As a consequence:
+ - Client loggers are now instance-specific by default, to ease debugging.
+ - End-users and developers can now configure logging as they wish, notably
+ to write down results to files.
+
+[33mcommit fda19ea5d986b8c97ae8b48ada3881387df60095[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Oct 10 12:08:43 2022 +0200
+
+ Disable time-out on websockets keep-alive pings.
+
+ * Websockets (optionnally, by default) uses keep-alive ping/pong
+ messages to prevent connections from being deemed idle and thus
+ closed e.g. by network proxies.
+ * Previously, declearn had been written in the intention not to
+ use this feature, however it was indeed active, and made open
+ connections to shut down under a (short) timeout due to ping/
+ pong apparently not working properly, even on the localhost.
+ * This commit revises the code to enable keep-alive ping/pong,
+ but disable timeout associated with them (hence the requests
+ are attempted, can hopefully bring stability to connections,
+ but will not disrupt declearn in case they fail in spite of
+ the connection still being active).
+ * Note: the actual future solution will be to implement proper
+ authentication and therefore reconnection mechanisms.
+
+[33mcommit eeae8c670d9dd91e5b65288311bff6471c0711ee[m
+Merge: 435c2c5 049ddb3
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Oct 5 10:07:23 2022 +0000
+
+ Merge branch 'optimize-model' into 'main'
+
+ Improve Model and its subclasses.
+
+ See merge request magnet/declearn/declearn2!2
+
+[33mcommit 049ddb32ef761fa5c742c488992c2c630b6fb266[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Wed Oct 5 10:07:22 2022 +0000
+
+ Improve Model and its subclasses.
+
+ Clean-up Model and its subclasses:
+ * Improve documentation of Model methods.
+ * Harmonize `_unpack_batch` method name for the current Model subclasses.
+ * Harmonize loss-averaging behavior of the current Model subclasses.
+
+ Improve TensorflowModel:
+ * Optimize gradients computation in `TensorflowModel` using `tf.function`.
+ * Improve the management and serialization of `loss` in `TensorflowModel`.
+ * Implement utilities to build keras Loss objects from strings or functions.
+
+[33mcommit 435c2c5326144d2913ecd344f34e3b129270ae92[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Oct 4 12:15:06 2022 +0200
+
+ Remove useless caching instructions from CI/CD config.
+
+[33mcommit 697dc900a7888344a5f287fab961741ae3b91c2b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Oct 4 12:09:38 2022 +0200
+
+ Add some metadata to installation files.
+
+[33mcommit 8877b95c16d7f84c5d01452a057bafad9049d473[m
+Merge: 376c878 b004fc8
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Oct 4 10:06:59 2022 +0000
+
+ Merge branch 'refactor-effort-constraints' into 'main'.
+
+ See merge request magnet/declearn/declearn2!1
+
+[33mcommit b004fc823bfadc98de1b508ae4388210596261cd[m
+Author: ANDREY Paul <paul.andrey@inria.fr>
+Date: Tue Oct 4 10:06:59 2022 +0000
+
+ Refactor and improve effort constraints applicable to clients.
+
+[33mcommit 376c8785b3a1ff4c157511ed95840182bbf75ff0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 29 17:15:47 2022 +0200
+
+ Minor docstring fix.
+
+[33mcommit 2f3c16a0a0df6ceab02ecdbbf68783a20dd1ec3e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 26 11:10:02 2022 +0200
+
+ Protect `TorchModel.compute_batch_gradients` outputs from side effects.
+
+[33mcommit 1726bae1b9199c6729131e026773e1c560b3f3fa[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 26 11:09:08 2022 +0200
+
+ Refactor `TorchModel` backend loss computation.
+
+[33mcommit 7ff16191ef4ba432022fa6a9c7651f644b6bb595[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 22 11:38:54 2022 +0200
+
+ Specify 'localhost' as '127.0.0.1' in tests to avoid CI failure.
+
+[33mcommit 2cd6995fdd85d8b445978e2e80e6bdd93f000394[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 17:01:52 2022 +0200
+
+ Avoid `test_torch[RNN]` failure in py38.
+
+[33mcommit c7a431e1688b16da681eee7e0d25656969ec8f11[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 15:33:43 2022 +0200
+
+ Add CI/CD tags to use shared inria runners.
+
+[33mcommit 7a1bca48345ffce294d6226b4dca82c07e070c89[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 15:07:51 2022 +0200
+
+ Revise CI/CD configuration.
+
+ Changes:
+ * Add some setup instructions, that should enable caching
+ installed packages rather than downloading them on each
+ and every pipeline call.
+ * Configure tests to run without '--fulltest' and on the
+ main branch only.
+ * Configure tests to run with '--fulltest' when a merge
+ request is created.
+
+ Rationale:
+ * Avoid wasting resources on CI/CD jobs on _every_ commit:
+ trust developers to test their code when relevant.
+ * Ensure tests are run on the main branch as it is somewhat
+ more critical than development ones.
+ * Automate tests in the context of MR, which is nice.
+
+[33mcommit 1bf8fc837e5545e85f5d3434d2225567e45b4554[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 14:56:33 2022 +0200
+
+ Add `--fulltest` commandline option and skip some tests by default.
+
+[33mcommit 6688db7e5934e9a0430c06f7d4eb3a0894ed50ed[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 14:29:28 2022 +0200
+
+ Use validation datasets in `test_main.py`.
+
+[33mcommit f1106ec2e477b64e5a5af82a4b94193a26093fa0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 14:14:42 2022 +0200
+
+ Fix registration timeout use in tests.
+
+[33mcommit d5f6200b76f18e2c95d360b8e7f0bfadac18a849[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 14:13:06 2022 +0200
+
+ Add default value for `RegisterConfig.min_clients`.
+
+[33mcommit 844b74100cd31436b7e0d96fb7ca972456dfeb5e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 12:56:28 2022 +0200
+
+ Add GitLab CI configuration file.
+
+[33mcommit 84aef330fdaa0f69c5962916fd29110128e4a433[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 12:55:31 2022 +0200
+
+ Add some internal cross-references to README.
+
+[33mcommit 14881b9dc7a33e1ceb8005aa359b7917205b93a6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 12:51:02 2022 +0200
+
+ Add `tox.ini` configuration file.
+
+[33mcommit bcef11d9edf2c6a6decbf273c780feb9620a8a30[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 12:50:38 2022 +0200
+
+ Update `pyproject.toml` file.
+
+[33mcommit 296f8045fa1f1ff28104d0b92f4442e761d4087e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 11:45:10 2022 +0200
+
+ Silence tensorflow-raised deprecation warnings.
+
+[33mcommit e77f9a352d2577984253aff61cdf912a1f121481[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 11:05:59 2022 +0200
+
+ Improve GrpcServer-GrpcClient non-xfail unit tests.
+
+[33mcommit 7ef961cc76c593a0ac8a851b953c2fc3913dcc55[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 11:05:10 2022 +0200
+
+ Add timeout for clients registration in tests.
+
+[33mcommit 21b1b48bc4f4c3ccb795dd27d98389e43972f2ce[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 21 10:03:01 2022 +0200
+
+ Fix and document optional dependencies' installation.
+
+[33mcommit 628c7142cb8f3bc75a7444699fcb3da1fea1b173[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 17:58:53 2022 +0200
+
+ Add a README file.
+
+[33mcommit ff4979a79db406c7298360629d2e12c4e0f3b0f5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 14:30:02 2022 +0200
+
+ Fix declearn imports in tests.
+
+[33mcommit a722a3cf59ec379275c5bc638101c37f2609e66b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 12:01:17 2022 +0200
+
+ Rename package to 'declearn'.
+
+[33mcommit c9068d1a63ceff853cc15afe32f3edb7f3ec4306[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 11:56:27 2022 +0200
+
+ Rename 'test' folder.
+
+[33mcommit 47dd4c968674594e945e962de9851bc467b89834[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 11:40:43 2022 +0200
+
+ Improve `test_routines.py` failure detection.
+
+[33mcommit e46b4f4307a1d3d2b6ee23b12737a87e9cd20490[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 11:31:42 2022 +0200
+
+ Fix `grpc.protobufs` submodule import.
+
+[33mcommit ac4f533b697f077c632bb219c8ae0a8e5fd2e753[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 10:49:00 2022 +0200
+
+ Update description and version in installation files.
+
+[33mcommit 916494342ec55341922eee14e7b0cde77dc19d63[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 10:46:25 2022 +0200
+
+ Add main package docstring.
+
+[33mcommit 31a97c920da5d139342fc15a3cd378a6af731234[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 10:26:55 2022 +0200
+
+ Fix a docstring typo.
+
+[33mcommit ab67e177f9b47cbfde26a9dcd3a410f9eedc3cfd[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 10:26:47 2022 +0200
+
+ Modularize and document `FederatedClient`'s init parameters.
+
+[33mcommit 302e0588468ba898a1bae42f0e59054d1ef55d9e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 10:06:49 2022 +0200
+
+ Implement generic `load_dataset_from_json`.
+
+[33mcommit 118ea9536275fa655019014bde163c7f0cebb9b0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 17:12:07 2022 +0200
+
+ Improve code readability of batch-generation kwargs' handling.
+
+[33mcommit 18c1ebc31d158b356667dbe752ef223e6cb7b511[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 17:10:26 2022 +0200
+
+ Deploy `Checkpointer` use in `FederatedClient`.
+
+[33mcommit a8d68bac42e8f495ca32a50b7bd2a9ecd0e9d4be[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 16:36:25 2022 +0200
+
+ Fix `FederatedServer._keep_training`.
+
+[33mcommit dcb53b431131c949f4da66702c00b4e1699208bf[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 16:15:07 2022 +0200
+
+ Ensure checkpoint folder is created.
+
+[33mcommit a947855196ece3dc12a28ef9e2564b705d350284[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 16:14:42 2022 +0200
+
+ Redistribute `FederatedServer` parameters between init and `run`.
+
+ Deploy `RegisterConfig` as part of this code refactoring.
+
+[33mcommit 212514e3ddfc400c6fdd6831051d648626b78b72[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 16:12:50 2022 +0200
+
+ Add `RegisterConfig` dataclass.
+
+[33mcommit a1d350c2a38649c28702d8d2c28e1ac964c8ebd6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 14:22:43 2022 +0200
+
+ Modularize FL training and evaluation parameterization.
+
+ * Implement dataclasses utils to wrap and parse training
+ and evaluation hyper-parameters.
+ * Expose and use these as part the `FederatedServer` API.
+ * Adjust information passed through messages.
+ * Adjust `FederatedClient` code to make use of the former.
+
+[33mcommit d73b65aa77b54179b6fed2b9ae4f1b44e1967ff8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 14:17:46 2022 +0200
+
+ Modularize and document `Server` init parameters.
+
+[33mcommit d1a6f9fc30830e28af0d4a943b01754badfe5291[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 14:01:26 2022 +0200
+
+ Rename `EarlyStoppingCriterion` to `EarlyStopping`.
+
+[33mcommit 702801013f381a2f06d9381e10cc283e5a2ceebc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 19 11:42:44 2022 +0200
+
+ Add dataclasses to wrap and parse communication endpoint config dicts.
+
+[33mcommit 657306bb3167f6e8c90b543a5f0446be1bb7ee8d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Sep 16 15:31:20 2022 +0200
+
+ Add checkpointing to "main" functional tests.
+
+[33mcommit 2753b2810a797e1329c66e28deedcc505445ef9c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Sep 16 15:22:18 2022 +0200
+
+ Implement server-side model and loss checkpointing.
+
+[33mcommit 019608b0e5c4c518aef26376d88c00bdba4b0832[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 15 10:53:15 2022 +0200
+
+ Fix `SklearnSGDModel` loss computation.
+
+[33mcommit dd288a0896a79cb3b74177f2182a9fe4162e19e5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 14 17:27:42 2022 +0200
+
+ Correct loss aggregation in `FederatedServer`.
+
+[33mcommit e3dc7bc36e3a0a1f0b1df035c68aeaacf1edfb7c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 14 17:20:32 2022 +0200
+
+ Correct `SklearnSGDModel` loss computation.
+
+[33mcommit 32bfca67a1a6e76435ba02a5186d3d213623c13f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 14 16:42:50 2022 +0200
+
+ Add `declearn.main` to load-time imports.
+
+[33mcommit b7200729a8b0404bd99e5133d4dd7cacdac5a2e1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 14 16:41:50 2022 +0200
+
+ Add optional validation data to `FederatedClient`.
+
+[33mcommit 10b401b2b1a16d3009aec264c9d8a6ef4b01af67[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 14 14:27:31 2022 +0200
+
+ Fix `SklearnSGDModel` initialization.
+
+[33mcommit 0a7d455013982963b35137fa4c63b8c2e3c4867f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 14 13:33:17 2022 +0200
+
+ Fix various bugs in `InMemoryDataset`.
+
+[33mcommit 4efea666431cbf5c891b7b6f1f42e9f4629731b8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 14 13:32:43 2022 +0200
+
+ Fix Yogi optimization module's formula.
+
+[33mcommit ad5c4e0fd61e2871a5a50c42b5e8502378556f51[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 13 15:27:58 2022 +0200
+
+ Extend documentation of communication `Client` and `Server`.
+
+[33mcommit adc08808cd7ff22951427a2e4ea16e76baf807d3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 13 12:38:11 2022 +0200
+
+ Remove unrequired `time.sleep` safeguards from `test_main.py`.
+
+[33mcommit 2b37a4b706ee5616ee617758cfcb2e9cc0396028[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 13 12:30:23 2022 +0200
+
+ Fix `test_grpc.py` unit tests.
+
+[33mcommit 4353965a324b65ecc52edbc529a5ab9ca3603d57[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 13 12:14:03 2022 +0200
+
+ Fix `GrpcClient.register` to catch connection failures.
+
+[33mcommit 03eb72398ea6794ab0e497fecdf32e35a092f200[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 13 12:13:18 2022 +0200
+
+ Add `WebsocketsClient` reconnection upon registration failure.
+
+[33mcommit c81105fb664f696f816bca8844942ef7ba1b80b4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 13 12:12:39 2022 +0200
+
+ Add 10-seconds reconnect attempts to `WebsocketsClient.start`.
+
+[33mcommit 104ac195518d98c73eb8f52dff3ee0a6defef6aa[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 13 12:11:18 2022 +0200
+
+ Expand `Client` start/stop methods' specifications.
+
+[33mcommit 94db32c6af190255b1584f55644a3111479eb5cd[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 17:12:21 2022 +0200
+
+ Implement functional tests for declearn's main classes.
+
+[33mcommit 5c55524e3bad42e69eba642163b6346e8c74e0e5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 16:38:17 2022 +0200
+
+ Fix `aggregate_data_info` edge case `required_fields=set()`.
+
+[33mcommit 0ab4c461a33b9be9739b1f55ce6b16023c39c7a5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 13:55:59 2022 +0200
+
+ Add JSON support for built-in set objects.
+
+[33mcommit 217553f0600d354687fb5098b56b9a92743d41d3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 13:51:28 2022 +0200
+
+ Prevent `InMemoryDataset.classes` from storing `numpy.generic` objects.
+
+[33mcommit e85d6ba8de6159be7ece5c239bf80a707a795428[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 13:40:09 2022 +0200
+
+ Remove port shifting from `communication/test_routines.py`.
+
+[33mcommit 42d1de49be9a131c7dfd841047fac03bf2c7293c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 13:34:11 2022 +0200
+
+ Improve auxiliary variables' specs in messaging.
+
+[33mcommit 32f367b0e436fc743c82f04be71c7179186d852d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 13:33:42 2022 +0200
+
+ Fix auxiliary variables' passing.
+
+[33mcommit 521faea4bbbfa73acc6411e8ec3bcaf66edcace2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 13:18:03 2022 +0200
+
+ Fix `FederatedClient.stop_training`.
+
+[33mcommit 89888ed3884c8cfaf6a8e4713d200e8dfe05237e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 12:57:30 2022 +0200
+
+ Deploy `EarlyStoppingCriterion` to `FederatedServer`.
+
+[33mcommit c88577bdf88d9d21ce4e5b6f2d701dc219eb5f4b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 12:33:46 2022 +0200
+
+ Implement `EarlyStoppingCriterion`.
+
+[33mcommit fadb322a178e944ff514753f02b2e7aed96aeab5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 12:33:01 2022 +0200
+
+ Refactor some code into `declearn2.main.utils` submodule.
+
+[33mcommit 68b1e50bd23d65f6bcd5c3a006cfc6071a4c160b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 12 12:30:44 2022 +0200
+
+ Improve documentation of FederatedServer data-info parsing code.
+
+[33mcommit 2ea3dccfdbfa167141b3e305398c12a43af50bb9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Sep 9 15:32:09 2022 +0200
+
+ Implement `stop_training` routines in main classes.
+
+[33mcommit 6b1be7544fbba7c2b3364365af7df2b2af51975d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Sep 9 14:02:08 2022 +0200
+
+ Fix type-hinting of exchanged model weights.
+
+[33mcommit f9317e55c67a904bf8145cdc2a9a4043cbf7eee8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Sep 9 13:35:32 2022 +0200
+
+ Implement functional tests for communication Client/Server use.
+
+[33mcommit caccedddc3249b7535b3aba03e52db8b0b85e7f4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Sep 9 11:31:23 2022 +0200
+
+ Implement loss evaluation rounds in main classes.
+
+[33mcommit 8b8e2ccb574c20abcba9d355a4ff1223d7c76d10[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 8 17:27:18 2022 +0200
+
+ Fix optimizer auxiliary variables' parsing on server side.
+
+[33mcommit ac9883e498a2bfb4f74a0ee1dff88452e78cc478[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 8 16:39:13 2022 +0200
+
+ Improved verbosity of main classes during initialization.
+
+[33mcommit 7d4c3f25dde6e02f03c91354d8a95ad8fac2a59d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 8 16:38:43 2022 +0200
+
+ Add `wait_for_clients` arguments to `FederatedServer.run`.
+
+[33mcommit 516428bb75732aaaa87961089185e658cd6bb306[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 8 16:37:03 2022 +0200
+
+ Fix `FederatedClient` initialization success reporting.
+
+[33mcommit 71a85279078d24f839f57896f205c1a1d5642758[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 8 16:36:28 2022 +0200
+
+ Fix communication messages' de-serialization.
+
+[33mcommit 6d1200865d7b9431f865376c9b1fe4f7b4da40dc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 8 15:43:32 2022 +0200
+
+ Add `expose_classes` parameter and attribute to `InMemoryDataset`.
+
+[33mcommit 762f9fe8c9669e2f116b95f6e6c9b70c36badc6d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Sep 8 15:43:04 2022 +0200
+
+ Fix `SklearnSGDModel.from_parameters`.
+
+[33mcommit e3a4954b294cfc0009cf2649f26a1c3d6acae2ac[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 16:47:36 2022 +0200
+
+ Deploy `Client` revision to `FederatedClient`.
+
+[33mcommit 12879d84e71f2e8dc5c863bcf80c57bfb9924527[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 16:45:34 2022 +0200
+
+ Revise and improve communications `Client` API.
+
+ * Modify `Client` in a similar way as `Server`:
+ - turn into an asynchronous context manager
+ - make `start` and `stop` methods asynchronous
+ - get rid of `loop` parameter and attribute
+ - add `_setup_ssl_context` abstract method to have
+ the parent class handle children' shared arguments
+
+[33mcommit 19efade6c6922ce2b9011d33cc5c8b2fb4486625[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 16:29:22 2022 +0200
+
+ Deploy API revisions to fix `FederatedServer`.
+
+[33mcommit fae428cd95fff501935bd5985a0ee11aa28cd860[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 16:28:48 2022 +0200
+
+ Revise and improve communication `Server` API and subclasses.
+
+ [API]
+ * Make `start` and `stop` methods asynchronous, enabling to get
+ rid of `self.loop.run_until_complete(...)` calls.
+ * Turn `Server` classes into asynchronous context-managers, so as to
+ automate `await server.start()` and `await server.stop()` calls.
+ * Remove `loop` argument and parameter completely from `Server`.
+ * Add `_setup_ssl_context` abstract method to `Server` in order to
+ delegate all shared arguments' processing to super in children
+ classes.
+
+ [GrpcServer]
+ * Refactor `MessageBoardServicer` subclass to ease readability.
+ * Delegate gRPC server creation to `start` method out of API
+ consistency, and to enable re-using the same `GrpcServer`
+ object multiple times.
+ * Make '8765' the default port (as with `WebsocketsServer`),
+ so that the runtime port-selection becomes a documented but
+ non-default feature (as it is unpractical for URI sharing).
+
+[33mcommit b44b1acdbd15bdfc45cce2bcce723a508c9b7b8a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 15:06:15 2022 +0200
+
+ Fix `declearn.communication.build_server`.
+
+[33mcommit 5b9ab599363a38376f61089d11b21ac74f69774a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 14:57:18 2022 +0200
+
+ Implement TorchModel (de)serialization.
+
+ Note: unfortunately, this relies on pickling the model,
+ which is unsatisfactory from a security perspective but
+ is the simplest practical way available at the moment.
+
+[33mcommit 10f9195b27a92c1b76c3629da0c53e1792c2111e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 13:17:07 2022 +0200
+
+ Avoid `asyncio.get_event_loop()` DeprecationWarning (Py>=3.10).
+
+[33mcommit 2daf688e90f7c0e981421587b63ceb6c088c8364[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 13:08:31 2022 +0200
+
+ Fix OptiModule unit tests.
+
+[33mcommit 305c7c09a0363d057e4918843db33ab754b12ff5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 12:05:03 2022 +0200
+
+ Add `compute_loss` method to the `Model` API.
+
+[33mcommit 8a9808949ef9b619f93c5dbbf2951ca7fd09c253[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 11:59:30 2022 +0200
+
+ Implement `TorchModel.compute_loss`.
+
+[33mcommit c12b25fa57b124bcc25e0c28ee441355ff64a505[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 11:59:12 2022 +0200
+
+ Implement `TensorflowModel.evaluate`.
+
+[33mcommit ebcd160616a02680c2d441c87b3e10c2c8762b00[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Sep 7 11:58:45 2022 +0200
+
+ Ensure proper training/evaluation behaviour is used with TensorFlow.
+
+[33mcommit 926255830a2b193b6bfa13014974703978c9e4ed[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 6 15:10:05 2022 +0200
+
+ Improve documentation of FederatedClient.
+
+[33mcommit cbe33f1f0ae062839ebe028fbb6b0753c47f5a73[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 6 12:50:11 2022 +0200
+
+ Update dependencies in declearn2 installation files.
+
+[33mcommit bd3130f831329cb5f0a4e20a0246a779d4d8ed96[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 6 12:20:44 2022 +0200
+
+ Improve documentation of FederatedServer.
+
+[33mcommit 4110d2bb7607572de2ea3e13665d2b83a3ddf701[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 6 12:20:23 2022 +0200
+
+ Enable partial clients' participation in training rounds.
+
+[33mcommit 406a1cdda8336178ce996e8de72c80c44da53b34[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 5 17:17:49 2022 +0200
+
+ Modularize clients' training round effort.
+
+[33mcommit 2e597c4f5b92df67aea0cef669eec638d041a8bb[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 5 16:41:12 2022 +0200
+
+ [WIP] Revise "main" classes' current rewrite.
+
+[33mcommit 98bde8718381d7c4cfc6a081df83d06d15215530[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 5 16:40:40 2022 +0200
+
+ Add `CancelTraining` message class.
+
+[33mcommit 58c172543b7ba0a61d8bf303d7a7a315ae508db5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 5 15:46:13 2022 +0200
+
+ Rewrite communication submodule.
+
+[33mcommit 5dd1883078685d9f6debb37ee429bbd3dd0c5acf[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Sep 5 15:32:56 2022 +0200
+
+ Fix internal imports in `strategy` submodule.
+
+[33mcommit 3185243b179677b32dc9c173aaad7de2f55ff15a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 17 18:47:52 2022 +0200
+
+ [WIP] Start integrating "main" classes.
+
+[33mcommit 62bb504349f45949443c938b1aecbc1dba58e5e0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 16 18:48:43 2022 +0200
+
+ Implement (scratched) new `Strategy` API and subclasses.
+
+ Notes:
+ * This commit has *not* been properly tested yet and will
+ most likely be patched soon.
+ * The current API is not to be considered stable~definitive
+ (although it may be kept for a while in order to finalize
+ the base re-integration of the main FL classes).
+
+[33mcommit c3575ab8b4e864d59369f81518d267e6ab284618[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 12 17:42:34 2022 +0200
+
+ Improve OptiModule documentation.
+
+[33mcommit b96bba042b816254b48e71ba4e6f972c47b5a33b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 12 17:25:00 2022 +0200
+
+ Correct Scaffold algorithm correctness.
+
+ * Rewrite local states' update:
+ - Simplify the formula to remove unrequired terms.
+ (previous version was correct but less effective)
+ - Raise in case state is set to be collected while
+ no optimization steps have been taken (otherwise
+ aggregation becomes wrong on the server side).
+ * Correct server-side behaviour: remove undue operation.
+ - No correction is to be applied on the server side to
+ the global updates; previous version was wrong.
+ - Thus the server-side module merely aggregates local
+ states and sends back information to all clients.
+ * Correct associated unit tests.
+
+[33mcommit a8f6e149106ec5ed7071260f5d016149713b65d6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 12 15:27:04 2022 +0200
+
+ Add `__repr__` methods to Vector subclasses.
+
+[33mcommit a59fcf1aa349fe3daf21a5d419a4f08a0234dc16[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 12 15:10:17 2022 +0200
+
+ Added type-registration to Model classes.
+
+[33mcommit d4a27aaef3eb46e28c01e35af7a268b0b49edd72[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 12 11:48:29 2022 +0200
+
+ Implemented 'data_info' requirements and initialization in Model API.
+
+ Note: such a mechanism exists in declearn1, which is thereof restored
+ and expanded thanks to the new `declearn.data_info` module.
+
+[33mcommit 052acf5b607a2f323b42fc83b669cd2b35859dec[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 12 11:26:46 2022 +0200
+
+ Add `InputShapeField` data_info specification.
+
+[33mcommit ee951ec80f0b8cc804d1ec0be9e6a628aa633fda[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 12 09:48:38 2022 +0200
+
+ Patch some docstrings in `data_info`.
+
+[33mcommit eb57526f3afe59d784a40900fe78d87b65164d54[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 11 18:17:49 2022 +0200
+
+ Implement `data_info` module to specify metadata fields.
+
+[33mcommit adb42ab76cd44bc7d0af1282887edddfc99326fc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 11 15:27:31 2022 +0200
+
+ Remove unrequired pylint comments.
+
+[33mcommit 4286a758a2204076da2a6b1f5ed1cf7cd612047a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 11 10:03:36 2022 +0200
+
+ Fix docstring typos.
+
+[33mcommit 885a479aef0c40c3e04a2efc6592f6b65cd0c3c5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 11 10:03:10 2022 +0200
+
+ Split some `test_grpc.py` code into `conftest.py` for re-use.
+
+[33mcommit 3685f242c3b0dcc29207de19529bcd8a4e0d2f54[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 11 10:02:16 2022 +0200
+
+ Fix type-registration of communication subclasses.
+
+[33mcommit 6cef948fd47e6baf1af681c6c2359822cf0a4db2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 10 17:22:58 2022 +0200
+
+ Add exception in `GrpcClient.ping` when client is stopped.
+
+[33mcommit 027ee266bbc5ac937b124e8a2194ca64a6414a9a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 10 17:22:22 2022 +0200
+
+ Add revised tests for low-level gRPC communications.
+
+[33mcommit cf7333efc2a4dd18c6eaa31eb40e41ee5733a5b3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 10 13:33:59 2022 +0200
+
+ Fix SSL setup of `GrpcServer`.
+
+[33mcommit b9470bd58742982919c3bbff3b7644cd537922f8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 10 11:30:56 2022 +0200
+
+ Restore `GrpcClient.ping` method.
+
+[33mcommit 05467ab1a9bac217bbc951c9fd83db13e59e542f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 10 10:51:15 2022 +0200
+
+ Add `client_names` property to `Server` classes.
+
+[33mcommit af3a5020882ad35e8daa52b0c415817a6188c1e2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 10 10:39:35 2022 +0200
+
+ Add client name de-duplication in `WebsocketsServer`.
+
+[33mcommit d41806daf03bb4507d152dd07cd45ff747f257a6[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 10 10:11:48 2022 +0200
+
+ Correct `GrpcServer.wait_for_clients`.
+
+ The method previously returned the full `{ip: client_dict}` dict
+ stored by the dependent service, while the API specifies that it
+ should return `{name: data_info}`. This commit enforces that.
+
+[33mcommit 5ef8b9ddd5153f538aa7864be8b733ca81caa03b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 10 10:08:38 2022 +0200
+
+ Add `send_message` async method to `Server` classes.
+
+ Rationale: the current `broadcast_message` method only enables
+ sending the _same_ message to all clients, whereas it is often
+ required to send distinct messages, be it to limit participants
+ to a training round, or to pass individual auxiliary variables.
+
+[33mcommit 68dcfa3c7af681aa8084290fa56f249e8d6fa67c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 18:40:27 2022 +0200
+
+ Enhance communications submodule's docstrings.
+
+[33mcommit 8affb88adad835f4fc4ff289679dc51e314f5bd0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 18:31:33 2022 +0200
+
+ Implement `build_client` and `build_server` utils.
+
+[33mcommit 2eb9d0062915c45c6562e25c035b6cac59ffa3cf[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 18:28:30 2022 +0200
+
+ Split `communication.api._base.py` into client and server files.
+
+ Note: this does not change _anything_ from the user's perspective
+ or to imports outside the submodule, as the classes are still
+ re-exposed at the `api` submodule level.
+
+[33mcommit e0049dcf73c457766ad2d9593bb00810f1d7f9b2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 17:56:12 2022 +0200
+
+ Revise gRPC backend code to improve clients' identification.
+
+ * Add client names' aliasing in Service class, enabling client
+ names' de-duplication (while preserving the initial info).
+ * Use requests' provenance to identify clients, and retrieve
+ their associated alias on the server side.
+ * Remove newly-unrequired 'name' attribute in some messages.
+ * Note: this is probably not a definitive design, as we might
+ want to enable clients' address to change and prefer a token-
+ or key-based authentication and identification method.
+
+[33mcommit 1e5354e66dfcb58d7b093166bbaa05021be3c857[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 16:52:58 2022 +0200
+
+ Patch some logger-related code in communication classes.
+
+[33mcommit 84fa86973022724419b096bfdfb474a1d4a43a09[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 16:51:30 2022 +0200
+
+ Add `uri` parameter to communication `Server` classes.
+
+[33mcommit 3bc1252b7ae3e3df38a56c40cfc78af6b0c01433[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 16:39:04 2022 +0200
+
+ Revise gRPC server-side backend code.
+
+[33mcommit 3b237e1011695a2df0df351c3486973eb398934a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 16:38:46 2022 +0200
+
+ Fix `GrpcServer.wait_for_clients` logging.
+
+[33mcommit 50e1e0f9536c4dae9614e243b302e1cf065dce51[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 16:02:06 2022 +0200
+
+ Abstract some code into base communication `Client`.
+
+[33mcommit f5bd3628549449c4e37b4073268aea3b013c5b41[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 15:57:53 2022 +0200
+
+ Integrate revised `GrpcClient`.
+
+[33mcommit ae972360094e8dd63644b9644edf128a9cd7ae0e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 15:53:13 2022 +0200
+
+ Patch gRPC server-side code on minor points.
+
+[33mcommit 723ebb2f18cfce83de6e934894c97fcd38c89444[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 15:21:30 2022 +0200
+
+ Integrate revised `WebsocketsClient`.
+
+[33mcommit 38058ac28b3d480208f8f01c11d11a312f8816f2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 15:20:55 2022 +0200
+
+ Revise communication `Client` API.
+
+[33mcommit 0d7554289a9d8a5c7be41a42fc485d01d618ddab[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 15:18:48 2022 +0200
+
+ Deploy `get_logger` to communication classes.
+
+[33mcommit 7ebb4201500fe08fa55c2d8239c91d794d16768a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 15:18:11 2022 +0200
+
+ Implement `get_logger` under `declearn.utils`.
+
+[33mcommit f65da040b8e46b933eb11924c0d95689f3c90a1b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 12:00:00 2022 +0200
+
+ Abstract some code into `Server` and harmonize children classes.
+
+[33mcommit 3c74516e2a082e6819dd0764801c7753ce790273[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 11:58:55 2022 +0200
+
+ Fix websockets imports tracking and rename WebsocketsServer class.
+
+[33mcommit 5ba88238bb6efae796af30549dbb650d79a65447[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 9 10:47:21 2022 +0200
+
+ Improve and harmonize password support in Server classes.
+
+[33mcommit 6fcccbe2cbcc36145bfb3e65c729d42efa217656[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 5 21:01:50 2022 +0200
+
+ Integrate revised gRPC server-side code.
+
+[33mcommit 314418a7ce007554bffa84560b799fa6fd131330[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 5 21:00:15 2022 +0200
+
+ Revise marginally the previous commit.
+
+[33mcommit 67ed275de0f2f7bcb4607bcda0953ddb5b611f92[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 5 16:52:56 2022 +0200
+
+ Start integrating revised communication into declearn2.
+
+[33mcommit 5fd46ded5c205c5c96de3378acaa23d6bf9deca8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Aug 5 11:25:22 2022 +0200
+
+ Revise docstring sections' formatting.
+
+[33mcommit 77ca114359a85dd38fe1b058d5455e333df797c3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 11:02:32 2022 +0200
+
+ Add `from_parameters` class constructor to `SklearnSGDModel`.
+
+[33mcommit c0565275250e536ef3e13a8f6d6d98779b1140ce[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 4 14:58:14 2022 +0200
+
+ Refactor Model subclasses' test code.
+
+[33mcommit b7692f960634e0edc7ef0612c0c11cde30ea916b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 4 13:43:05 2022 +0200
+
+ Temporarily replace 'Self' type annotations in Vector.
+
+ * Self should be the proper way to annotate return types
+ for Vector that are in fact subclass-dependent so that
+ they are properly type-checked by static linters.
+ * However, as of the current (0.971) version of mypy, it
+ is not yet well-linted, causing a lot of "type: ignore"
+ comments in code.
+ * For the moment, we therefore drop Self from Vector base
+ code (but preserve it in concrete implementations). As
+ soon as mypy supports Self, this will be changed back.
+
+[33mcommit 5faabbfddbf939130c69efc0df8205f1caa477cc[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 4 13:22:49 2022 +0200
+
+ Fix some type annotations in `NumpyVector`.
+
+[33mcommit 29dabfdc86c438f46634b461f0c6057e5069cdaa[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 4 11:46:24 2022 +0200
+
+ Fix unit tests' linting issues (code-only changes).
+
+[33mcommit fb86fdf5bbe34e5bb5251a9242a7c9252946fdbf[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Aug 4 11:30:54 2022 +0200
+
+ Revise Vector building and serialization API.
+
+ * Replace `Vector.__new__` type-inference trick with `Vector.build`
+ staticmethod to ease code readability (as well as type checking).
+ * Extend the behaviour of `register_vector_type`: in addition to the
+ backing of `Vector.build`, this decorator now registers decorated
+ classes under the 'Vector' group name, and most importantly adds
+ JSON (de)serialization support based on the object's `pack` and
+ `unpack` methods.
+ * Note: among the three actions taken by `register_vector_type`,
+ the last one (JSON support) is probably the most desirable.
+ * Modify models' unit tests to take these changes into account;
+ the gain in concision in gradients' (de)serialization tests'
+ code is a direct result from those changes.
+
+[33mcommit fc7f91334a541c67e2ad9ea3b21591c52ae1ebad[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 19:58:01 2022 +0200
+
+ Implement unit tests for Scaffold optimization modules.
+
+[33mcommit 91b916b49ff043e7d77fb866bdb4a95c470a1506[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 19:56:45 2022 +0200
+
+ Remove Vector serialization from Scaffold modules.
+
+[33mcommit 99c607df414b7b365aff5efb747aefd59e137142[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 19:31:38 2022 +0200
+
+ Fix `ScaffoldClientModule` state-update computation.
+
+[33mcommit 36a154ad5b317e31d610de7f5ad7a75e09ee02ec[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 19:30:09 2022 +0200
+
+ Fix `ScaffoldServerModule.collect_aux_var`.
+
+[33mcommit 403a7a848f33ea99c00c5ceb2149133b94f5ac8e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 19:28:34 2022 +0200
+
+ Fix un-optimized equality checks in Vector subclasses.
+
+[33mcommit 56b94b231e73f323395d13254607ecaaddc6d644[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 18:32:51 2022 +0200
+
+ Replace `Vector` (de)serialize methods with (un)pack ones.
+
+ * The rationale behind this change is to delegate serialization
+ to classes that handle Vector instances while providing a way
+ to define Vector-subclass-specific conversion to/from a JSON-
+ serializable format (e.g. numpy thanks to declearn utils).
+ * This is not entirely satisfactory though and will probably be
+ further modified soon (before integration in declearn).
+ * Another point to alter is the "register_vector_type" function,
+ which could be removed in favor of "register_type" (in which
+ case (un)pack might become (get|from)_config, unless overall
+ serialization utils get revised as well).
+
+[33mcommit ab26a1ec368c215039a177528561ce21b508c47f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 18:10:19 2022 +0200
+
+ Refactor and correct Vector backend code.
+
+ * Expose additionnal public and private methods at Vector
+ base class level.
+ * Add private-attributes-based operator overriding, using
+ stdlib `operator` to provide working dunders at Vector.
+ * Use Self type-hint and `type(self)` statements to make
+ code both abstract and applicable. Note: mypy does not
+ properly type-check 'Self' yet but should do so in the
+ upcoming months as it is integrated into Python 3.11.
+ * Remove a lot of redundant code in children classes.
+
+[33mcommit e03047ea437304a01e95d2fc7fa59141b99bd1d8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 14:37:19 2022 +0200
+
+ Implement generic unit tests for `OptiModule` subclasses.
+
+[33mcommit 53e52b5738b3a4c31549a1b85ed7fdcff2a69011[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 14:28:15 2022 +0200
+
+ Correct `ScaffoldClientModule.collect_aux_var` behaviour.
+
+[33mcommit 33dcd88a907f77e98c7eaba533cadea2e03cec85[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 11:33:58 2022 +0200
+
+ Implement unit tests for `utils._serialize` submodule.
+
+[33mcommit 8b3d45cc9e22d42a5833df8eab6562810efdce0d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 10:22:07 2022 +0200
+
+ Add `get_config` / `from_config` methods to `Optimizer`.
+
+[33mcommit 8292dcd44458c6615b5416789d54d7ce03c7937b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 10:21:39 2022 +0200
+
+ Improve `declearn.utils._serialize` tools' documentation.
+
+[33mcommit 371a566ee6c1e2ad977a6be9edb2fa0ded8fea5f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Aug 3 10:21:16 2022 +0200
+
+ Add support for dict inputs to `deserialize_object`.
+
+[33mcommit cb8a55210f41c6c4f87eef52f37b3b848b51ae73[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 17:58:30 2022 +0200
+
+ Deployed config-based (de)serialization to `OptiModule` classes.
+
+[33mcommit f6bb094b7820a367eadc7ddcc47dffe3da76e88e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 17:40:39 2022 +0200
+
+ Silenced tensorflow-import warnings in unit tests.
+
+[33mcommit 75e09c3364846e8ea03de86e8c52d28793cb29de[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 17:37:13 2022 +0200
+
+ Implement unit tests for `utils._json` submodule.
+
+[33mcommit ccaf835d9feca2e14a05e2e6155608e1e289ee30[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 16:39:49 2022 +0200
+
+ Implement unit tests for `utils._register` submodule.
+
+[33mcommit fdea21267e0de114032f7f998efcc598fbd90d00[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 16:00:43 2022 +0200
+
+ Implement (de)serialization tools for `SupportsConfig` objects.
+
+[33mcommit 24eafb34b5f70ed0ff4f7b7b2998bcdde6f429e4[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 15:59:32 2022 +0200
+
+ Repackage and register numpy-arrays-(un)packing utils.
+
+ * Changes to the private import structure do not alter
+ users' access to the implemented functions.
+ * numpy arrays are now supported in JSON dump/load calls
+ provided the proper functions are used a object hooks.
+
+[33mcommit 76a9d49b1cf55eee17327633b73bc79bac07eae2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 15:55:33 2022 +0200
+
+ Implement JSON (un)packing utils.
+
+ * Define and make public `add_json_support` to enable adding support
+ for arbitrary types' (un)packing in JSON load/dump calls.
+ * Define and make public `json_pack` and `json_unpack` functions,
+ which are to be deployed in declearn's backend JSON load/dump
+ calls and ensure registered (un)packing protocols are used.
+
+[33mcommit ad6c650ebc15b3b0c8fb8ecf8d2ba7e59ac7954b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 12:02:15 2022 +0200
+
+ Reorder `from typing_extensions import ...` statements.
+
+ Although it provides with 'future' elements from the `typing` standard
+ library, `typing_extensions` is a third-party package. Statements were
+ therefore re-ordered to abide by PEP-8.
+
+[33mcommit 202b4f6e535176281884dbb19e592330160d739c[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 11:56:44 2022 +0200
+
+ Implement `SupportsConfig` Protocol-type.
+
+[33mcommit ba7c798d91b4f9abf4a6074acf6953835a4d3290[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 11:45:16 2022 +0200
+
+ Update installation files.
+
+ * pyproject.toml is now compatible with setuptools'
+ toml-only installation; in the future, setup.cfg
+ (and legacy setup.py) may therefore be removed.
+ * toml-only install with setuptools does not (yet)
+ support editable install though, hence setup.cfg
+ is preserved for now (for declearn2 installation)
+ * At the moment, installing declearn2 without the
+ installation file for declearn interferring is
+ somewhat difficult; this will be resolved when
+ declearn2 revisions are integrated into the main
+ package.
+
+[33mcommit 0347605a7ce07002684265d4df83aac587738bfe[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 10:49:28 2022 +0200
+
+ Fix tests' compatibility with Python 3.7.
+
+[33mcommit 2a0c5167c6aa5f674940cd6a1f8325eed1285d08[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Aug 2 10:48:56 2022 +0200
+
+ Correct setup.cfg
+
+[33mcommit 1c8d1129afe7ccb0f27a1b91e4a7572186f07e19[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 18:03:47 2022 +0200
+
+ Implement reverse-lookup types-registration utils.
+
+[33mcommit 7b7e8ced71643ec5755206d7d7fbce04eb362b95[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 17:14:11 2022 +0200
+
+ Deploy types-registration to OptiModule classes.
+
+[33mcommit 8e980649ea1b68fa6bfa78ff5f20fb5639d80682[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 17:10:36 2022 +0200
+
+ Implement types-registration utils.
+
+[33mcommit a29a7053a5250629a381eb27b2e26ac7dabdae8a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 16:59:56 2022 +0200
+
+ Refactor `utils` submodule into a folder structure.
+
+[33mcommit 0a675009fc8ed0f675748ac960217e55de046c5b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 16:45:42 2022 +0200
+
+ Improve docstring formatting of some optimizer modules.
+
+[33mcommit 09020dc64fd1b5c7a9182ba5bcaf2709c046ad3d[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 16:42:54 2022 +0200
+
+ Update main submodules' import rules.
+
+[33mcommit 2dfba8e9aceca550c6f0e6bcb61248a75fd4a9b5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 16:41:34 2022 +0200
+
+ Implement `Optimizer` API and base class.
+
+[33mcommit 40aea29075e41b1612f2806efe8668ab49935e2e[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 16:41:14 2022 +0200
+
+ Improve and document `OptiModule` API.
+
+[33mcommit 03e3cc76706e2fd0f22496abc6b10ada3f28e495[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 14:39:14 2022 +0200
+
+ Implement SCAFFOLD as a pair of `OptiModule` classes.
+
+ Note: OptiModule and overall Optimizer APIs
+ need further consolidation for SCAFFOLD to
+ be effectively working, but the core of the
+ algorithm and its documentation are here.
+
+[33mcommit 8cfda044ad6ea5c590d46758d35526e96d647a28[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Mon Aug 1 14:34:53 2022 +0200
+
+ Trivial changes to backend optimizer modules' files.
+
+[33mcommit f3b475b156c3ba365d8fdc655a3543ca91e0d5ef[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 17:30:09 2022 +0200
+
+ Repackage some code into `declearn.optimizer.modules`.
+
+[33mcommit cac92d9ae2df1c4676868408f64686984df46c4f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 16:37:00 2022 +0200
+
+ Implement PyTorch unit tests.
+
+ Note: the tested neural network architectures
+ and declearn functionalities are the same for
+ PyTorch and TensorFlow save for features that
+ are still being implemented.
+
+[33mcommit 587cbc0010966b059fc2893b7ab9b50d8ae94d06[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 16:35:36 2022 +0200
+
+ Alter TensorFlow unit tests marginally.
+
+[33mcommit c9be2facea395d3351249c47eb2e7e7274b77b65[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 16:35:17 2022 +0200
+
+ Fix `TorchModel`'s handling of numpy inputs.
+
+[33mcommit b62af968dd520e77e517112f4ab0e3ba708856ff[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 16:34:52 2022 +0200
+
+ Fix `TorchModel.apply_updates`.
+
+[33mcommit afd659a3fff476b251d0716f05d951ebb8d189f2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 16:34:33 2022 +0200
+
+ Fix `TorchModel` get/set weights.
+
+[33mcommit dc84469ecb33d5b90d3003c3c0bd8a9ca34bc8b7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 16:34:00 2022 +0200
+
+ Fix `TorchVector.deserialize` warning.
+
+[33mcommit eec7f413317dd1fef0fb6b3b61f6d2efcf784877[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 14:41:04 2022 +0200
+
+ Add NumpyVector gradients (de)serialization unit test.
+
+[33mcommit ff28fd40dfe694bfc91a13c7e724eaa89b06fcf0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 14:40:45 2022 +0200
+
+ Add `TensorflowModel.compute_loss`.
+
+[33mcommit a764735f86d8b5b3542d1fbd64b9ae33ec7b8a00[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 13:29:26 2022 +0200
+
+ Implement unit tests for TensorflowModel (and thus TensorflowVector).
+
+[33mcommit 9ded427582535aa2853645cc6128befb820114d8[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 13:28:40 2022 +0200
+
+ Harmonize `apply_updates` formula between SklearnSGD and Tensorflow.
+
+[33mcommit da39226f7461983172f31967eb4c395ae965b5c5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 13:28:08 2022 +0200
+
+ Add numpy-inputs support for `TensorflowModel`.
+
+[33mcommit f46c16259389c0edf2014db867ec9ea05b5e8d32[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 13:27:04 2022 +0200
+
+ Add full support for `tf.IndexedSlices` in `TensorflowVector`.
+
+[33mcommit d2f0754ad2c54e27bd58b0c09fc2030a34f9d4d9[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 10:57:23 2022 +0200
+
+ Improved SklearnSGD tests' documentation.
+
+[33mcommit 9bb0ef2a1c0deaa22e0d841ed8f4ddacf580f1d3[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 10:26:59 2022 +0200
+
+ Refactored SklearnSGD tests to make them unitary.
+
+[33mcommit fa22131ee2bede0391410a1de257df01c165faab[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Fri Jul 29 10:26:14 2022 +0200
+
+ Improved SklearnSGD test parametrization.
+
+[33mcommit e3f530f130518fe3c4d4607436acc794e9976810[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 28 16:52:10 2022 +0200
+
+ Lighten optimization algorithms' documentation.
+
+[33mcommit 5d354188be017148beb87c72e88f1222c77335c1[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 28 16:51:15 2022 +0200
+
+ Turn off the automatic import of `tensorflow` and `torch` submodules.
+
+[33mcommit 44b6215f059a562624b5ed5903e88a2e4ffcf511[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 28 16:50:59 2022 +0200
+
+ Rename `SklearnDataset` to `InMemoryDataset`.
+
+[33mcommit 2bb61b785373e1001ea3ec36d4c8b80daf668171[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 28 16:50:49 2022 +0200
+
+ Implement unit tests for `SklearnSGDModel`.
+
+[33mcommit e1bbe5050b3091bff0c192c06b1ef755a326fbf2[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 28 14:35:02 2022 +0200
+
+ Fix `SklearnSGD` (de)serialization and classifier initialization.
+
+[33mcommit 9b867d01c6f521d324dfeb0e4cc7e3844cd6d1a7[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 28 11:31:28 2022 +0200
+
+ Add updated installation files for declearn2.
+
+ * Use pyproject.toml / setup.cfg files, as `python setup.py install`
+ has been deprecated for some time now.
+ * For now, install `declearn2` rather than `declearn` for revised API
+ testing purposes.
+ * Note that as `setup.py` exists, it will have side effects on the
+ installation. This will be resolved once revisions are merged into
+ the actual library.
+
+[33mcommit 337dc951d885d8c4f94c5deaeffc4423354781e0[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 28 11:26:05 2022 +0200
+
+ Revise data batch specification and deploy new `Batch` type-hint alias.
+
+ * Change the specification of data batches in the `Dataset` and `Model`
+ APIs: a batch is now a (x, y, w) tuple, where y and w may be None.
+ * As a side-effect, remove some boilerplate code and set `unpack_batch`
+ util for future garbage collection as it should now be useless.
+ * Implement a `Batch` type-hinting alias to ease code readability.
+ * Place the latter in a new `typing` submodule (as is done in numpy).
+
+[33mcommit bf6bbef2ba15f323d3410858f6e6c75743c4444b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Thu Jul 28 11:16:23 2022 +0200
+
+ Add sample_weights support to `SklearnDataset`.
+
+[33mcommit d992051b35634d3ca6960c0f2ace40750b667965[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 17:54:40 2022 +0200
+
+ Implement `SklearnSGDModel.compute_loss`.
+
+[33mcommit d3ba4c56e8d0ebed91366eb7d5eba16577c4f086[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 17:53:57 2022 +0200
+
+ Fix SklearnSGDModel for regression models.
+
+[33mcommit 8f55af2913ba7776b6e94b87de8f078b36601b35[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 16:13:00 2022 +0200
+
+ Fix SklearnSGDModel for classifier models.
+
+[33mcommit 1b30b51c7eb7561701534b2b856fb0e041a8406f[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 15:00:20 2022 +0200
+
+ Correct TorchVector deserialization.
+
+[33mcommit e4c6ecc1de23abeb89d2b414a22bb82eb3169548[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 15:00:02 2022 +0200
+
+ Improve Vector subclasses' minimum and maximum methods.
+
+[33mcommit 454f93268c308b124bcceb6824374f390c37133b[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 13:46:27 2022 +0200
+
+ [WIP] Scratch optimizers' modular implementation.
+
+[33mcommit 6ed543704a8d9fc7a55930fae02b399a3bfb1f4a[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 13:29:36 2022 +0200
+
+ Scratch the implementation of Dataset API.
+
+[33mcommit fbd3d397604f8b5079780803b48389496ed78332[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 13:27:58 2022 +0200
+
+ Implement PyTorch Model and Vector interfaces.
+
+[33mcommit 6e3230a6cb9618cd7c0b3e3a4ff89ab97c44feee[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 13:27:40 2022 +0200
+
+ Implement Tensorflow Model and Vector interfaces.
+
+[33mcommit c581633fc83732010ef82534d9dc8b221e646cd5[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Wed Jul 27 13:27:11 2022 +0200
+
+ Rewrite SklearnSGDModel.
+
+ * Made a single class to wrap either a classifier
+ or regressor model.
+ * Abided by the design implemented at Model level.
+ * Note: classifier might need additional work for
+ proper initialization.
+
+[33mcommit d4f10c0c5a1b90d559fb480a6a45f66a214b1cad[m
+Author: Paul Andrey <paul.andrey@inria.fr>
+Date: Tue Sep 20 10:58:45 2022 +0200
+
+ Rewrite Model and Vector API code.
+
+ * Changed Vector to be a container for multiple
+ data structures by design.
+ * Skimmed and type-hinted Model API.
+ * Implemented generic NumpyVector container.
+ * Added (de)serialization functionalities to
+ the Vector API and its numpy implementation.
--
GitLab