Integrating FairXplainer with AIF360

[bishwamittra]

Two files are added to integrate FairXplainer.

1. aif360/sklearn/explainers/bias_explainer.py
2. examples/sklearn/demo_bias_explaination.ipynb

(1) contains the interface to call FairXplainer from AIF360 and (2) provides a notebook demonstration.

[bishwamittra]

Hi @mnagired,

I have added descriptions of explanations according to your suggestion. Hope this addresses your question.

Best regards, Bishwa

[mnagired]

@hoffmansc, this PR looks good to me. Please have a look whenever you get a chance :)

[bishwamittra]

Hi @hoffmansc

I have addressed your comments. The summary of changes are:

1. pytest added for bias explanation.
2. FairXplainer returns explanation results without bias value. Bias values can be computed by existing metric in aif360, as demonstrated in the notebook for statistical parity and equalized odds. However, I have not been able to compute predictive parity (sufficiency metric) using aif360. Therefore, I have added an optional argument "return_bias", which is by default False. If "return_bias=True", the bias computed by FairXplainer is additionally returned.
3. Other comments are addressed to my best efforts.

Thanks, Bishwa