Fix train script
* Check point callback was initialised with on_best_model callback which overwrote the model on the end of each epoch, even when the valuation loss did not improve * Fix passing on_best_model callback only if cfg.SEND_BEST_MODEL is set in the config * Delete model.save() on_best_model as model saving is performed by Keras base class of MyCPCallback.
Please register or sign in to comment