Using Keras fit_generator to train the model

It might be useful to use generator, see example here to train the model.

We should also add callback_list while training e.g. ModelCheckpoint, EarlyStopping, ReduceLROnPlateau to fit_generator function in order to save the best model (divide training set into training and development set).

I can work on it and send the PR over the weekend 😃

[rkcosmos]

notebook is still running on my local computer, but other small changes are done. So you can implement fit_generator and call_back on top of current code. I think fit_generator can be structured similar to this (https://github.com/rkcosmos/Breast-Cancer/blob/master/src/inception1_simplified.ipynb). We would need a function like ‘organise training set section’ to separate train and test folder. Then fit_generator can do its job naturally.

[titipata]

I think I’ll close the issue for now and we can take into account the fit_generator later once the training data is much larger (e.g. we want to incorporate Wikipedia). The version that we have currently using fit is pretty easy to read.