implement abstract automatic training manager

AllenNeuralDynamics / aind-behavior-curriculum

Starter repository for behavior base primitives.

https://aind-behavior-curriculum.readthedocs.io

MIT License

1 stars 0 forks source link

implement abstract automatic training manager #13

Closed dyf closed 1 month ago

dyf commented 4 months ago

no database stuff, use pd.DataFrame as your interaction layer.

hanhou commented 4 months ago

Example: https://github.com/AllenNeuralDynamics/aind-foraging-behavior-bonsai-automatic-training/blob/ef3ee668ad0532d509a7e61160d68df7b68c15a5/code/aind_auto_train/auto_train_manager.py#L27

dyf commented 3 months ago

@jwong-nd, notes from 3/26:

update to ensure that multiple policies can be active at once
@bruno-f-cruz said something about tuple[Policy,Stage] "is for analysis" that I didn't understand
need to more extensive test cases (backwards transitions, manual transitions)
do we need policy constraints? need to discuss.

bruno-f-cruz commented 3 months ago

Basically make sure that the history output structure is formalized somewhere instead of a generic Tuple object. I suspect that people might want to use this for analysis.

jwong-nd commented 2 months ago

Recently updated with multiple active policies, see bae0d32

jwong-nd commented 2 months ago

Adding final tests in branch 'trainer-tests'

jwong-nd commented 2 months ago

Testing complete

bruno-f-cruz commented 1 month ago

This is done, closing.