Fix `TensorflowModel` gradients' labeling

Closes #14 (closed) (see that issue for details on what is being fixed here)

Tasks:

• update TensorflowModel.compute_batch_gradients to use variables' names to label the returned tensors
• consider updating TensorflowModel.apply_updates to run some verifications on inputs' names
• extend Model unit test suite to include a verification that gradients and weights share the same specifications

added bug label

assigned to @paandrey

changed title from Draft: Resolve "[BUG] TensorflowModel weights and gradients have distinct labels." to Draft: Fix TensorflowModel gradients' labeling

changed the description

added 1 commit

• c55a3265 - Fix the labeling of output gradients in `TensorflowModel`.

Compare with previous version

marked the checklist item update TensorflowModel.compute_batch_gradients to use variables' names to label the returned tensors as completed

marked the checklist item consider updating TensorflowModel.apply_updates to run some verifications on inputs' names as completed

added 1 commit

• 261ab379 - Add gradients-labeling check to the Model test suite.

Compare with previous version

marked the checklist item extend Model unit test suite to include a verification that gradients and weights share the same specifications as completed

The commits pushed until now address the issue identified in issue #14 (closed).

They do not fix issue #15 (closed), which also identifies cases when the misalignment of gradients and variables might break the optimizer computations. Since this issue is more general and requires a minor API extension, I would like to merge the present MR, and open a dedicated one for issue #15 (closed) afterwards.

requested review from @nbigaud

approved this merge request

marked this merge request as ready

approved this merge request

enabled an automatic merge when the pipeline for 261ab379 succeeds

merged

mentioned in commit e8afbe1b