[features] Add pipeline
Opened this issue · 7 comments
okbalefthanded commented
- evaluation pipeline
okbalefthanded commented
- Cross validation strategies available:
- Kfold (only for SingleSubject evaluation)
- Shufflesplit (for both types of evaluation: Single/Cross subject)
okbalefthanded commented
Long running evaluations may face interruptions, a checkpoint mechanism is required to resume operations from interruption point
- Evaluation checkpoint and resume
okbalefthanded commented
Add Cross dataset evaluation for domain adaptation
- CrossSet
okbalefthanded commented
- Hyperparameter tuning with Keras-tuner
okbalefthanded commented
- Experimental:
- Add preprocessing layers before any model
- Break pipeline : delete preprocessing methods and leave it to the model
okbalefthanded commented
- Set image date format in the evaluation for model's training/inference consistency on GPU/TPU/CPU.
okbalefthanded commented
- Add Debug mode to evaluation:
- Keras Tensorboard callback.
- TF Debugging V2