Naive Bayes Classifier with Mixed Bernoulli/Gaussian Models

Description

I suggest allowing mixed datasets (half binary variables, half real-valued variables) into the Naive Bayes classifier. Currently the GaussianNB and BernoulliNB classes handle one case or the other, but not combined. I’d be happy to write the code for this, so I’m curious if this has been explored before and if it would be helpful!

For example, on the Titanic dataset, gender is a bernoulli variable while age is real-valued. Passing both into a Naive Bayes classifier would improve it.

This is related to this currently pending PR: https://github.com/scikit-learn/scikit-learn/pull/12569

[gautam-e]

I think a Mixed / General Naive Bayes classifier that allows one to mix and match the already available sklearn implementations of Naive Bayes with categorical and continuous columns is a use case that has been left unattended. It has been asked several times on Stackoverflow etc. with mostly pretty unsatisfactory answers. I have already derived and written some code that does this but have never made a PR before. Would be happy to team up on this @jarednielsen

[FlorianWilhelm]

@jarednielsen Are you working on this? As @jnothman already said, you could also start to work on a GeneralNB even if the CategoricalNB is not merged right now. Let @timbicker and me know if you need any pointers or want to discuss something. We would like to support this endeavour.