[FEA] Multi-class classification in SVM

[tfeher]

Is your feature request related to a problem? Please describe.
Currently only binary classification is supported.

Describe the solution you'd like
Implement multi-class classifier.

Additional context
Watch out for parameter layout #958

[beckernick]

This would be awesome addition 👍

[alxhrzg]

Just for someone looking for and intermediate solution, you can wrap e.g. SVC in sklearns OneVsRestClassifier wrapper such as:
For some background:

• Nesting classifiers in OneVsRestClassifier

from sklearn.metrics import make_scorer
from sklearn.multiclass import OneVsRestClassifier
from sklearn.metrics import accuracy_score

accuracy_typed = make_scorer(lambda y_true, y_pred: accuracy_score(
    y_true.astype(int), y_pred.astype(int)
), greater_is_better=True)


SVC_params = {
    'estimator__C': [0.1, 1, 10, 100, 1000],  
    'estimator__gamma': [1, 0.1, 0.01, 0.001, 0.0001], 
    'estimator__kernel': ['rbf']
}  

SVC_Grid = GridSearchCV(
    OneVsRestClassifier(SVC(verbose=2)), 
    SVC_params, 
    refit = True, 
    verbose = 3,
    scoring=accuracy_typed
)