History of keras fit() in when fitting multiple times

user_16425179 0
Viewed 232

This question is basically an extension of the issue raised in the Github forum of the keras repository:

https://github.com/keras-team/keras/issues/4446

The question was whether it is possible to fit() a keras model multiple times (e.g. due to a very large training set).

An alternative would be to train on batches, but let's stick to the above mentioned case.

If I would train such a model and retrieve the history after the last iteration of the loop, would it be the history of the last fit() iteration or a history over all fits (e.g. a mean of the losses at each epoch)?

Thank you in advance for your answers!

keras tensorflow history python
1 Answers

Closing this thread since the answer is already given (to some extent) in the discussion I linked to that post https://github.com/keras-team/keras/issues/4446

with the answer of OverLordGoldDragon on Jun 7, 2019.

He mentions that the history has to be saved in the loop, hence most likely the history after the final loop is just of the last fit() iteration,

user_164251790
Related