Numerical features

[alexeygrigorev]

Currently it supports only binary features. It would be nice to support numerical features as well, so feature weighting schemes (such as TF-IDF) could be used.

Right now there is already an implementation in a separate branch (numerical), but there's no single interface that supports both types of model.

It would be nice to have the ability to specify type on model creation. Something like

model = ftrl.FtrlProximal(alpha=1, beta=1, l1=10, l2=0, features_type='binary')

or

model = ftrl.FtrlProximal(alpha=1, beta=1, l1=10, l2=0, features_type='numeric')

[alexeygrigorev]

Also, it is not clear how much overhead it gives to use numerical features. Maybe it's ok to just use them all the time and leave it to the user to use binary features - by using 1.0 as feature weights