Adding SHAP predict values as new output file

[mattahrens]

New output file shap_values.csv in xgboost_predictions folder will contain feature importance for prediction set.

[amahussein]

Thanks @mattahrens
The csv file shap_values.csv is missing in the PR

[mattahrens]

The shap_values.csv is generated at runtime in the xgboost_predictions folder, so I didn't think it needed to be checked in

[amahussein]

@cindyyuanjiang Please take a look at this PR as you were adding the CLI to see if there is anything that affects the CLI feature.

[leewyang]

@mattahrens LGTM. Keep in mind that this prints the shap values across the entire dataset. If you want to see per-sample values, then you'd need to modify this line, but then you'd also end up with a dataframe of shape (num_samples, num_features) vs (num_features, 1) currently. However, I'm not sure that would be very usable vs. debugging via individual shap waterfall plots offline.

[mattahrens]

I wanted to start with the values across the entire dataset and get feedback. And then if we want to go per-sample, we can enhance it later to provide that.