scikit-learn
diff --git a/‎sklearn/linear_model/ridge.py
Lines changed: 11 additions & 10 deletions b/‎sklearn/linear_model/ridge.py
Lines changed: 11 additions & 10 deletions
@@ -961,8 +961,8 @@ class _X_operator(sparse.linalg.LinearOperator):
     """
 
     def __init__(self, X, X_mean, sqrt_sw):
-        self.n_samples, self.n_features = X.shape
-        super().__init__(X.dtype, (self.n_samples, self.n_features + 1))
+        n_samples, n_features = X.shape
+        super().__init__(X.dtype, (n_samples, n_features + 1))
         self.X = X
         self.X_mean = X_mean
         self.sqrt_sw = sqrt_sw
@@ -991,15 +991,16 @@ class _Xt_operator(sparse.linalg.LinearOperator):
     """
 
     def __init__(self, X, X_mean, sqrt_sw):
-        self.n_samples, self.n_features = X.shape
-        super().__init__(X.dtype, (self.n_features + 1, self.n_samples))
+        n_samples, n_features = X.shape
+        super().__init__(X.dtype, (n_features + 1, n_samples))
         self.X = X
         self.X_mean = X_mean
         self.sqrt_sw = sqrt_sw
 
     def _matvec(self, v):
         v = v.ravel()
-        res = np.empty(self.n_features + 1)
+        n_features = self.shape[0]
+        res = np.empty(n_features)
         res[:-1] = (
             safe_sparse_dot(self.X.T, v, dense_output=True) -
             (self.X_mean * self.sqrt_sw.dot(v))
@@ -1008,7 +1009,8 @@ def _matvec(self, v):
         return res
 
     def _matmat(self, v):
-        res = np.empty((self.n_features + 1, v.shape[1]))
+        n_features = self.shape[0]
+        res = np.empty((n_features, v.shape[1]))
         res[:-1] = (
             safe_sparse_dot(self.X.T, v, dense_output=True) -
             self.X_mean[:, None] * self.sqrt_sw.dot(v)
@@ -1119,8 +1121,9 @@ def _compute_gram(self, X, sqrt_sw):
         X_mean *= n_samples / sqrt_sw.dot(sqrt_sw)
         X_mX = sqrt_sw[:, None] * safe_sparse_dot(
             X_mean, X.T, dense_output=True)
-        X_mX_m = np.empty((n_samples, n_samples), dtype=X.dtype)
-        X_mX_m[:, :] = np.dot(X_mean, X_mean)
+        X_mX_m = np.full((n_samples, n_samples),
+                         fill_value=np.dot(X_mean, X_mean),
+                         dtype=X.dtype)
         X_mX_m *= sqrt_sw
         X_mX_m *= sqrt_sw[:, None]
         return (safe_sparse_dot(X, X.T, dense_output=True) + X_mX_m
@@ -1275,7 +1278,6 @@ def _solve_covariance_sparse_no_intercept(
         Used when we have a decomposition of X^T.X
         (n_features < n_samples and X is sparse), and not fitting an intercept.
         """
-        n_samples, n_features = X.shape
         w = 1 / (s + alpha)
         A = (V * w).dot(V.T)
         AXy = A.dot(safe_sparse_dot(X.T, y, dense_output=True))
@@ -1294,7 +1296,6 @@ def _solve_covariance_sparse_intercept(
         (n_features < n_samples and X is sparse),
         and we are fitting an intercept.
         """
-        n_samples, n_features = X.shape
         # the vector [0, 0, ..., 0, 1]
         # is the eigenvector of X^TX which
         # corresponds to the intercept; we cancel the regularization on