Change batchsize of 15_000 in NN_tools.py:evaluate_model to the batch size in training config
Currently the batchsize in NN_tools.py:model_evaluation and evaluate_model_umami are hardcoded to 15_000 for the model.evaluate and model.predict functions. This isn't a problem for DIPs and CADs which can train with such large batchsizes but can be problematic for larger models or inputs.
Required changes are needed:
- Pass train_config to the evaluate_model{,_umami} function, get the batchsize from that
- Modify MyCallback{,Umami} class to have an init function getting the training config to pass to these functions
- Modify calls to evaluate_model from calc_validation_metrics function