semi supervised da - correction + example

[Slasnista]

Hi,

I may have found a line that could lead to errors when using OT objects in a supervised DA setting.

at line 992 in da.py I propose to change classes = np.unique(ys) into classes = [c for c in np.unique(ys) if c != -1] which would enable people to use source samples with no labels to find the optimal couplin.

I also propose to add an example for semi supervised DA.

Do you agree with these propositions ? If yes, I'll open a PR.

[rflamary]

Hello @Slasnista

OK but be very clear in the documentation about what -1 means because this could introduce bugs for people who have 2 classes and use -1,1 as class.