remykarem / mixed-naive-bayes

Naive Bayes with support for categorical and continuous data

Oh hey didn't see you there. Here are some of my projects.

Utils

python2jupyter | pyterator

ML/DS

apriori-rs | cattrees | mixed-naive-bayes

Tensor/dataframe manipulation

pandas-lightning | autograd

Planning, search, reinforcement learning

py2pddl | visualise-dijkstra-shortest-path

Programming language

bython

Others

data-science-interview | url-shortener | The Kiasu Grocer

mixed-naive-bayes's Issues

PyPi API Documentation hyperlink not working

The link on PyPi for API Documentation section has not been updated and currently point to the URL that does not exist.

meaning for feature distributions

can you please clarify , what you mean

under feature distributions per :

discrete features that are categorically distributed. The categories of each feature are drawn from a categorical distribution.
https://scikit-learn.org/stable/modules/generated/sklearn.naive_bayes.CategoricalNB.html#sklearn.naive_bayes.CategoricalNB

seems to be you mean that distributions are estimated from data?
for given categorical feature it is probabilities and conditional probability (values and target) from calculated from data?

as mentioned in
https://datascience.stackexchange.com/questions/58720/naive-bayes-for-categorical-features-non-binary
Some people recommend using MultinomialNB which according to me doesn't make sense because it considers feature values to be frequency counts

Can't open documentation

Hey I tried to access the documentation but it gives me 404. Can you please fix it?

Question about code 268-278, p and t probability calculated by different NB and logic for normalizing them

@remykarem Hi Remy
Thank you for the fantastic work! I notice you calculated the probability for response to be in certain class p, t based on different NB, and tried to combine p, t to a final probability in code 268-278. Yet the usage of pt confused me a little, do they need to be weighted here? Extreme case would be, we have one categorical variable and rest 100+ continuous, pt will off-set the impact of continuous variable. Also could you provide more detailed guild in setting the self.priors?
Thank you very much for the sharing.
B.R.

predict_proba bug

Hello!

I hope you are doing well.
I'm learning Data Science and found your Mixed Naive Bayes implementation on pip. I tested the model and everything worked fine, until I noticed that the method predict_proba is not working correctly. The sum of the different probabilities for the different categories does not add up to 1.
I've discussed this situation with my colleagues and they all run into the same problem. Do you know if it is an error of the implementation of that method?
I am leaving attachments to show you the bug, and I also sent you an email with the Jupyter Notebook and the dataset in case you want to see in detail what I am talking about.

Thank you for the effort you put into building this model! If you happen to have a solution or/and explanation to the problem I ran into I would really appreciate it.

Good luck!

Integrating GMM

Hi @remykarem , thanks for this library.

Do you anticipate integrating mixture models rather than single Gaussians per feature? This has a lightweight implementation.

fit prints array of integers

Why does fit print an array of integers every time it is called? Is there a way to turn it off?

Sklearn interface

Hello,

I would have one suggestion/request. When designing scikit like interface, it would be actually extremely benefitial to implement scikit compliant classifier by inheriting from scikit base classes.

You can find how to do it here: https://scikit-learn.org/stable/developers/develop.html

The reason for this request is that sometimes the sklearn demands various properties (like classes_ property) which you can get for free by inheriting from their base classes.

Unfortunately, right now the model is not suitable for things like sklearn ensambling.

Bug report - priors validation

Hello,

I've found your lib on StackOverflow and found it awesome. There is just a little bug which makes it unusable when trying to play with the priors param.

When you do your validation that priors are 1, you are actually raise an error when they are valid.

Please see: https://github.com/remykarem/mixed-naive-bayes/blob/master/mixed_naive_bayes/mixed_naive_bayes.py#L156

The line 156, should actually be:

if not np.isclose(self.priors.sum(), 1.0):

May I ask you to fix it?

Anyway thanks for your time to put this lib together! It is really helpful :-)

David