Choose the minimal conditions to allow fit

[wdevazelhes]

In #127, we saw that algorithms should return a special error message if the number of samples is too small to be able to run. Therefore, we should choose the minimal number of samples on which we want to allow fit. If theoretically the algorithm would work with only n sample, we should ensure that the code indeed works with n samples as input. Therefore we should ideally also test these minimal conditions (minimally allowed conditions and forbidden conditions).

For instance: MMC_Supervised should theoretically work on two samples (either if they are from the same class, meaning MMC will put them into a point, or if they have a different class, in this case it can put them to any distance superior to 1)

Or also for instance maybe LSML_Supervised would at least need 4 samples to be able to form one quadruplet

[bellet]

(This should also be investigated for weakly supervised algorithms)

[wdevazelhes]

(This should also be investigated for weakly supervised algorithms)

Yes indeed