FIX: explicit conversion to original sparse format in SMOTENC (#539)

Author: glemaitre

doc/whats_new/v0.5.rst

@@ -31,3 +31,7 @@ Bug
   was moved before the activation function and the bias was removed from the
   dense layer.
   :issue:`531` by :user:`Guillaume Lemaitre <glemaitre>`.
+
+- Fix bug which converting to COO format sparse when stacking the matrices in
+  :class:`imblearn.over_sampling.SMOTENC`. This bug was only old scipy version.
+  :issue:`539` by :user:`Guillaume Lemaitre <glemaitre>`.

imblearn/over_sampling/_smote.py

@@ -815,6 +815,8 @@ def _sample(self, X, y):
 
             if sparse.issparse(X_new):
                 X_resampled = sparse.vstack([X_resampled, X_new])
+                sparse_func = 'tocsc' if X.format == 'csc' else 'tocsr'
+                X_resampled = getattr(X_resampled, sparse_func)()
             else:
                 X_resampled = np.vstack((X_resampled, X_new))
             y_resampled = np.hstack((y_resampled, y_new))