Does PNET support training with multiple classes?

[ilibarra]

Dear @marakeby,

Thanks for developing and releasing PNET. I am trying to do exploratory analyses in some multimodal data using this framework. Labels are more than two, so I wonder whether a multiclass flag is already available by default in one of the reported architectures tested in the paper. I think small changes in the nn.py and/or model_params_ dictionary could allow this, yet I have been testing without much success (e.g. 'prediction_outputto None, and a alternativeloss` key).

Do you think there are already defaults for trying this? Thanks for any comments on this.