Use multi_dot for Hessian and gradient product

jjerphan · jjerphan · commit 3b60347ef3d9 · 2021-02-26T14:55:14.000+01:00
np.linalg.multi_dot quickly chooses the best order for the multiplication of three matrices. See: https://github.com/numpy/numpy/blob/860c8b82939d1535351f7b651c284a55efe21b10/numpy/linalg/linalg.py#L2742 Changes to be committed:
diff --git a/sklearn/linear_model/_logistic.py b/sklearn/linear_model/_logistic.py
@@ -233,7 +233,7 @@ def _logistic_grad_hess(w, X, y, alpha, sample_weight=None):
 
     def Hs(s):
         ret = np.empty_like(s)
-        ret[:n_features] = X.T.dot(dX.dot(s[:n_features]))
+        ret[:n_features] = np.linalg.multi_dot(X.T, dX, s[:n_features])
         ret[:n_features] += alpha * s[:n_features]
 
         # For the fit intercept case.
