Support new multi-class objectives in lgbm_classification_learner

[IgorKhomyanin]

Status

READY

Todo list

• [x] Documentation
• [x] Tests passed
• [x] Issue: closes #222

Background context

The lgbm_classification_learner supported only a multiclass objective for a multi-class classification problem.

When a softmax (alias for multiclass) or a multiclassova (and all its aliases) were used as an objective, lgbm_classification_learner failed to produce a prediction function, resulting in an error.

Description of the changes proposed in the pull request

• Added a set LGBM_MULTICLASS_OBJECTIVES that contains all multi-class objectives that can be used in LGBM as of today.
• Created a boolean variable is_multiclass_classification that determines whether a multi-class problem is being solved or not.
• Replaced two instances of params["objective"] == "multiclass" with a single is_multiclass_classification boolean variable. This modification directs the flow inside the prediction function p to support new multi-class objectives, instead of just multiclass.

Where should the reviewer start?

Changes were only made inside lgbm_classification_learner.

[caique-lima]

Closing and reopening to trigger CI