Add public partial_fit to BernoulliRBM

[ogrisel]

BernoulliRBM is an online model but the incremental learning method is hidden in the private _fit API.

BernoulliRBM should have a public partial_fit akin to the one in MinibatchKMeans and MinibatchDictionaryLearning (along with tests and maybe an example using the digits dataset and random nudging to generate an infinite stream of samples while monitory the performance accuracy of a stacked linear model trained on a fix heldout training set and tested on a fix validation set).

[dsullivan7]

Does the private _fit method do what the partial_fit method should be doing? If this is the case, will the fit method make a call instead to the partial_fit method? I noticed that the partial_fit and fit methods for MiniBatchDictionaryLearning are essentially the same except for the batch_size parameter. Will this new partial_fit method mimic the fit method with a different batch_size set?

[ogrisel]

I have not checked in depth but _fit of BernoulliNB looks very similar to what a partial_fit should typically do. Whether fit should internally call partial_fit or a private _fit method also used by partial_fit is an implementation detail left to the developer. It mostly depends on whether or not there are costly redundant call that can be avoided by introducing a private _fit method.

[ogrisel]

Fixed by #2428.