FIX ZeroDivisionError in BaseLibSVM._sparse_fit when n_SV == 0 (scikit-learn#14894)

danna-naser · NicolasHug · commit a89462b59d1b · 2019-10-07T16:28:42.000-04:00
diff --git a/doc/whats_new/v0.22.rst b/doc/whats_new/v0.22.rst
@@ -572,6 +572,9 @@ Changelog
   :class:`svm.OneClassSVM` was previously non-initialized, and had size 2. It
   has now size 1 with the correct value. :pr:`15099` by `Nicolas Hug`_.
 
+- |Fix| fixed a bug in :class:`BaseLibSVM._sparse_fit` where n_SV=0 raised a
+  ZeroDivisionError. :pr:`14894` by :user:`Danna Naser <danna-naser>`.
+
 :mod:`sklearn.tree`
 ...................
 
diff --git a/sklearn/svm/base.py b/sklearn/svm/base.py
@@ -287,11 +287,14 @@ def _sparse_fit(self, X, y, sample_weight, solver_type, kernel,
         n_SV = self.support_vectors_.shape[0]
 
         dual_coef_indices = np.tile(np.arange(n_SV), n_class)
-        dual_coef_indptr = np.arange(0, dual_coef_indices.size + 1,
-                                     dual_coef_indices.size / n_class)
-        self.dual_coef_ = sp.csr_matrix(
-            (dual_coef_data, dual_coef_indices, dual_coef_indptr),
-            (n_class, n_SV))
+        if not n_SV:
+            self.dual_coef_ = sp.csr_matrix([])
+        else:
+            dual_coef_indptr = np.arange(0, dual_coef_indices.size + 1,
+                                         dual_coef_indices.size / n_class)
+            self.dual_coef_ = sp.csr_matrix(
+                (dual_coef_data, dual_coef_indices, dual_coef_indptr),
+                (n_class, n_SV))
 
     def predict(self, X):
         """Perform regression on samples in X.
diff --git a/sklearn/svm/tests/test_svm.py b/sklearn/svm/tests/test_svm.py
@@ -690,6 +690,19 @@ def test_sparse_precomputed():
         assert "Sparse precomputed" in str(e)
 
 
+def test_sparse_fit_support_vectors_empty():
+    # Regression test for #14893
+    X_train = sparse.csr_matrix([[0, 1, 0, 0],
+                                 [0, 0, 0, 1],
+                                 [0, 0, 1, 0],
+                                 [0, 0, 0, 1]])
+    y_train = np.array([0.04, 0.04, 0.10, 0.16])
+    model = svm.SVR(kernel='linear')
+    model.fit(X_train, y_train)
+    assert not model.support_vectors_.data.size
+    assert not model.dual_coef_.data.size
+
+
 def test_linearsvc_parameters():
     # Test possible parameter combinations in LinearSVC
     # Generate list of possible parameter combinations
