tf2 estimator and training strategies
Closed this issue · 3 comments
- train_sequence cannot be defined here but has to be used as defined in models/base/estimator.py
- tf2 training strategies have to be adapted as discussed
train_sequence
deleted
trainingStrategies.py
deleted and integrated into model/estimator.
All changes within train/tf2
.
@davidsebfischer If you checked whether your train_sequence
method is working with our train
method already, we are fine, otherwise I need to check that before we can close here.
train_sequence
calls train
multiple times, if more than one optimizer is given. However, each time train
is executed, a new model is created again in our case. So this isn't yet working, instead, the training will start from scratch for each given optimizer.
Please tell me, if I need to fix this before closing here.
Please check that the new global train_sequence works for you, then close here.
You can close this even if train sequence restarts every time but open a new issue for the train sequence please.
train_sequence is working fine now. We might need to include additional information into th Enums of training_strategies.py
in the future.