training_results

This folder contains folders with training results. Folder names corespond to folder names in /models. Additionally number of parameters and lowest log-loss on validation is appended to folder names.

Folder contents:

• model - hdf5 file with trained model (WARNING! This is end result, NOT the BEST result)
• model_weights - hdf5 file with weights from model file
• model.json - json description of architecture
• model_notlearned - model before first epoch
• model_weights.$EPOCHNUM.$VAL_LOSS.hdf5 - hdf5 files with weights. To get the best weights you have to get one of this files with lowest VAL_LOSS.
• on_epoch_end_acc.csv - csv file with correct classification percent (50% threshold) in learn set after each epoch
• on_epoch_end_loss.csv - csv files with log-loss in learn set after each epoch
• on_epoch_end_val_acc.csv - csv with correct classification percent (50% threshold) in validation set after each epoch
• on_epoch_end_val_loss.csv - csv files with log-loss in validation set after each epoch