Sequential.fit() validation_split argument - why not used?

Part 1 (2017)
1

Hello all!

Been reading through a lot of keras documentation and when looking at the fit() method in Sequential, I saw that there is a validation_split argument that takes a float (0 < x < 1) as a percentage of the data to split. Is there any reason there is a preference for globbing and creating valid directories instead of utilizing this method?

Thanks!

2

Well, I’m answering my own question here, but if I had the initial question, others may have it in the future, so here’s my rationale.

As mentioned in Lesson 3, Data Augmentation should always be used. This means that a generator should always be used as well. As a result, Sequential.fit_generator() will likely be used far more often than vanilla .fit().

.fit_generator() does not have a validation_split() method like .fit() does, so pre-defining the split would be necessary.

Hope this helps people out! :slight_smile:

3

thanks corbin, i was doing the same as you and would’ve asked the same question.