add notion of experiment campaign
While an Experiment
is the interface to a unique effort for training, we may add an additional (wrapping) notion/layer called an "(experiment) campaign". A campaign would define conditions in which one or more experiments will run.
Campaign would be both a researcher side (describing the campaign) and node side (describing the campaign and managing authorization for the campaign) notion.
Campaign definition may include (not exclusive):
- researcher X
- run model Y1/Y2/etc.
- on nodes A/B/C/etc. or subset of these nodes
- using dataset Z1/Z2/Z3/etc.
- with model testing policy (which datasets, local/global testing - see #184 (closed) )
- during time period T_begin to T_end
- experiment parameters: aggregator, strategy, training params, model params, breakpoints, etc.
- which/where results to save on researcher after experiment
- which results to transmit to nodes (or others ?) after experiment (see #228)
Goal is to better match hospital workflow.
Edited by VESIN Marc