Hi, suppose I train my model, by that time, I could see loss acc val_loss val_acc info, and I save it for later re-use, and comparison of result.
model = conv1(batches)
model.save_weights(model_path + ‘statefarm1.h5’)
While days after, I load the trained model,
model.load_weights(model_path + ‘statefarm1.h5’)
how could I get the previously loss acc val_loss val_acc info ? I means, once I retrain, the previous info in jupyter would loss.
Are these infos saved in statefarm1.h5, so I could later get them back ?
If not, what’s the recommended way of save these infos, and later re-get them back for comparison ?
If you do model.save(...) instead of save_weights(), it also stores the state of the optimizer. (But I’m not sure if that state includes the previous loss and accuracy metrics.)
With Keras, you can use a callback function with the fit generator and output the validation loss or validation accuracy directly in the saved filename
filepath = 'weights.{epoch:02d}-{val_loss:.2f}.model’
history = model.fit_generator(
# (…) initial parameters,
callbacks=[ModelCheckpoint(filepath, monitor=‘val_acc’, save_best_only=True)]
)
You can take a look at : https://keras.io/callbacks/
thanks, @alexandrecc
how to save the history object to a file ? reading through the source code, have not find this.
And the properties of history, I only see val_loss and val_acc, not found loss, and acc.