Error with modshogun

[wdevazelhes]

I think there is a problem when using shogun:
First after I installed shogun (the last version, I tried to install with conda forge and also pip), I cannot import modshogun (it throws a package not found error), but instead I can import shogun. Maybe I messed up somewhere in the install process or maybe the newest version have a different name or sth..

Even if I change the code with import shogun instead of import modshogun, there is another error when doing:

from metric_learn import LMNN
from sklearn.datasets import make_blobs
X, y = make_blobs(n_features=2)
X.shape
lmnn = LMNN()
lmnn.fit(X,y)

And it returns the following error:

NotImplementedError: Wrong number or type of arguments for overloaded function 'new_MulticlassLabels'.
  Possible C/C++ prototypes are:
    shogun::CMulticlassLabels::CMulticlassLabels()
    shogun::CMulticlassLabels::CMulticlassLabels(int32_t)
    shogun::CMulticlassLabels::CMulticlassLabels(shogun::SGVector< float64_t >)
    shogun::CMulticlassLabels::CMulticlassLabels(shogun::CFile *)
    shogun::CMulticlassLabels::CMulticlassLabels(shogun::CBinaryLabels *)

[bellet]

It is probably a version issue?

[perimosocordiae]

When I set up shogun initially, I built it from source and enabled the "python_modular" interface.

It seems like things have changed since then: shogun-toolbox/shogun#3898

We should update our usage of Shogun, or remove it entirely if it's going to be a pain.

[bellet]

Also it does not seem like Shogun_LMNN can do dimensionality reduction, which is at odds with the objective of #193. So maybe it is easier to remove it? We could try to work on an optimized version of LMNN later