different options for prediction

[rguliev]

In GitLab by @cbeleites on Jun 16, 2017, 13:21

In predict, allow the user to choose whether the result is normalized or not. Default should be to normalize.

Reason: while the standard formulation of the problem are mixtures, this is IMHO not necessary. At least, not always and not for the prediction. And there are several plausible ways to deal with predictions that do not have rowSum of 1 (and/or are outside [0, 1]).