Read/save bins Issue #96

[alexliap]

This PR aims to resolve Issue #96 by adding to class methods to OptimalBinning object, to_json & read_json.
Continuous and Multiclass binning inherit from that class, so no other methods were needed.
The to_json method also saves the transformation of the training data as requested in the Issue.

[guillermo-navas-palencia]

Hi @alexliap.

Thanks for taking the time to develop this PR. However, this is not what I had in mind (sorry if the description was lacking details). The json file should contain all the information needed to build the binning table.

[alexliap]

Thank you for the response.
I will work on it, but could you be a little more specific in order to avoid refactoring again?

[alexliap]

What about now, i save all the data required for the binning table, but is this the way you want it to behave? should the self._check_is_fitted() change in order to be able to call self.binning_table instead of self._binning_table ?

[guillermo-navas-palencia]

Yes, this looks good to me. Regarding self._is_fitted, yes, I think it should be set to True when reading from JSON.

[guillermo-navas-palencia]

To complete the PR, it would be nice to have unit tests for all binning classes. I mention this because the binning table classes require different arguments depending on the target type, so continuous and multiclass targets will not work automatically.

[alexliap]

Yes of course, I'm on it

[alexliap]

uuum question, why MulticlassOptBinning doesn't require dtype of variable as input?

[guillermo-navas-palencia]

uuum question, why MulticlassOptBinning doesn't require dtype of variable as input?

MulticlassOptBinning only accepts numerical dtype.

[guillermo-navas-palencia]

Thanks!