Support for Multinomial Naive Bayes

[krishnateja614]

Is there any estimate on when we can expect PMML support for Multinomial naive bayes model in sklearn?

[vruusmann]

What is the broader context of this feature request? The documentation for the sklearn.naive_bayes.MultinomialNB estimator type states that it's mostly suited for text classification purposes. However, the JPMML-SkLearn project doesn't address NLP workflows at the moment (eg. see issue #4), which means that there's no point in supporting the MultinomialNB estimator type as well.

[krishnateja614]

My problem consists of a column that takes many categorical variables (much
larger than the number sklearn random forest could handle natively which is
32 if I'm not wrong) and after one hot encoding the column, I was running
Multinomial Naïve Bayes which resulted a in very good accuracy compared to
other approaches. I guess this is similar to tf weighting where each
document contains only 1 word so a binary value instead of actual count.
Hence, I infact do not need any part of the NLP workflow like
countvectorizer or tfidf vectorizer but only Multinomial Naïve Bayes.

On Nov 22, 2016 1:46 AM, "Villu Ruusmann" notifications@github.com wrote:

What is the broader context of this feature request? The documentation for
the sklearn.naive_bayes.MultinomialNB estimator type states that it's
mostly suited for text classification purposes. However, the JPMML-SkLearn
project doesn't address NLP workflows at the moment (eg. see issue #4
#4), which means that
there's no point in supporting the MultinomialNB estimator type as well.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#20 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AHFGE4oVnYO8ZszsbOAjZ1P6du7UjhPmks5rAqv2gaJpZM4K45fq
.

[KaidongYu]

@vruusmann I saw that this project support CountVectorizer now, is there any plan to support Multinomial Naive Bayes & Bernoulli Naive Bayes in sklearn?

[vruusmann]

@KaidongYu I'm definitely going to add proper support for the NaiveBayesModel element into all my conversion libraries - JPMML-R, JPMML-SkLearn and JPMML-SparkML. I will be doing it all in one step, but I don't know when exactly do I have time for all that.

[Edzo0]

@vruusmann is it true that the Multinomial Naive Bayes si still not supported?