fixed multiprocessing issue during RoBERTa prediction - Solution to #1558

[etqadkhan]

With reference to #1558
When using a RoBERTa model for doing prediction, when the model is loaded using ClassificationModel() it prompts a warning stating,

UserWarning: use_multiprocessing automatically disabled as xlmroberta fails when using multiprocessing for feature conversion.

When prediction is performed on text data, if the number of records are a little more than a handful, the prediction progress bar takes infinite execution.

The issue occurs because in the classification_model.py file under the ClassificationModel class, there are two arguments that concern with multiprocessing, them being, args.use_multiprocessing and args.use_multiprocessing_for_evaluation. While one is set to False by default, the other remains to be True,as can be seen in the screenshot attached:

To be able to perform prediction by successfully disabling multiprocessing, we need to disable args.use_multiprocessing_for_evaluation = False and it should work fine. The approach has been tested locally and has proven to be working. Screenshot attached: