Add Ability to Convert Continuous Data to Categorical

mljar / mljar-supervised

Python package for AutoML on Tabular Data with Feature Engineering, Hyper-Parameters Tuning, Explanations and Automatic Documentation

https://mljar.com

MIT License

3.01k stars 401 forks source link

Add Ability to Convert Continuous Data to Categorical #117

Open eladmw opened 4 years ago

eladmw commented 4 years ago

Add Ability to Convert Continuous Data to Categorical : sklearn.preprocessing.KBinsDiscretizer

pplonski commented 4 years ago

It can be added for sure. But for now, it is hard for me to tell you how this can be exposed to the user. Options that I can see:

AutoML detects when feature after discretization help model achieves better performance and apply it
The user provides the info on which features need to be discredited. This is will probably need a new method in the interface.

Any ideas how would you like to see it?

eladmw commented 4 years ago

User info could be accepted. For example, if model = AutoML(), model.discretize(train,columns=[],bins=int,replace=False), where replace can replace continuous with categorical if wanted. It's also possible for this to be detected during the fit process and recommended to the user as an output. Also, I think that more categorical features would normally help because CatBoost looks for combinations of categorical variables to aid in boosting.