[dask] support 'pred_contrib' in predict()

[jameslamb]

Summary

LightGBM's Booster.predict() can be use to get feature contributions, by passing pred_contrib=True. This should be supported for model objects from the Dask interface as well.

Motivation

This is useful for inspecting the relative importance of features in a LightGBM model.

Description

The check at

LightGBM/python-package/lightgbm/dask.py

Line 169 in aae4fe4

if proba:

should also check whether pred_contrib=True has been passed, and return the full DataFrame of the result.

References

• this change in a fork of dask-lightgbm shows what the relevant change might look like: h2oai/dask-lightgbm@c19000b

[StrikerRUS]

Closing in favor of being in #2302 with other feature requests. Please leave a comment here if you'd like to work on this.

[jameslamb]

opening this as I'm actively working on it at the moment