scikit-learn
diff --git a/‎sklearn/manifold/_t_sne.py
Lines changed: 20 additions & 34 deletions b/‎sklearn/manifold/_t_sne.py
Lines changed: 20 additions & 34 deletions
@@ -654,18 +654,13 @@ class TSNE(BaseEstimator):
 
         .. versionadded:: 0.22
 
-    square_distances : True or 'legacy', default='legacy'
-        Whether TSNE should square the distance values. ``'legacy'`` means
-        that distance values are squared only when ``metric="euclidean"``.
-        ``True`` means that distance values are squared for all metrics.
+    square_distances : True, default='deprecated'
+        This parameter has no effect since distance values are always squared
+        since 1.1.
 
-        .. versionadded:: 0.24
-           Added to provide backward compatibility during deprecation of
-           legacy squaring behavior.
-        .. deprecated:: 0.24
-           Legacy squaring behavior was deprecated in 0.24. The ``'legacy'``
-           value will be removed in 1.1 (renaming of 0.26), at which point the
-           default value will change to ``True``.
+        .. deprecated:: 1.1
+             `square_distances` has no effect from 1.1 and will be removed in
+             1.3.
 
     Attributes
     ----------
@@ -755,7 +750,7 @@ def __init__(
         method="barnes_hut",
         angle=0.5,
         n_jobs=None,
-        square_distances="legacy",
+        square_distances="deprecated",
     ):
         self.n_components = n_components
         self.perplexity = perplexity
@@ -771,7 +766,6 @@ def __init__(
         self.method = method
         self.angle = angle
         self.n_jobs
10000
 = n_jobs
-        # TODO Revisit deprecation of square_distances for 1.1-1.3 (#12401)
         self.square_distances = square_distances
 
     def _fit(self, X, skip_num_points=0):
@@ -808,26 +802,19 @@ def _fit(self, X, skip_num_points=0):
             raise ValueError("'method' must be 'barnes_hut' or 'exact'")
         if self.angle < 0.0 or self.angle > 1.0:
             raise ValueError("'angle' must be between 0.0 - 1.0")
-        if self.square_distances not in [True, "legacy"]:
-            raise ValueError("'square_distances' must be True or 'legacy'.")
+        if self.square_distances != "deprecated":
+            warnings.warn(
+                "The parameter `square_distances` has not effect and will be "
+                "removed in version 1.3.",
+                FutureWarning,
+            )
         if self._learning_rate == "auto":
             # See issue #18018
             self._learning_rate = X.shape[0] / self.early_exaggeration / 4
             self._learning_rate = np.maximum(self._learning_rate, 50)
         else:
             if not (self._learning_rate > 0):
                 raise ValueError("'learning_rate' must be a positive number or 'auto'.")
-        if self.metric != "euclidean" and self.square_distances is not True:
-            warnings.warn(
-                "'square_distances' has been introduced in 0.24 to help phase "
-                "out legacy squaring behavior. The 'legacy' setting will be "
-                "removed in 1.1 (renaming of 0.26), and the default setting "
-                "will be changed to True. In 1.3, 'square_distances' will be "
-                "removed altogether, and distances will be squared by "
-                "default. Set 'square_distances'=True to silence this "
-                "warning.",
-                FutureWarning,
-            )
         if self.method == "barnes_hut":
             X = self._validate_data(
                 X,
@@ -907,7 +894,7 @@ def _fit(self, X, skip_num_points=0):
                     "All distances should be positive, the metric given is not correct"
                 )
 
-            if self.metric != "euclidean" and self.square_distances is True:
+            if self.metric != "euclidean":
                 distances **= 2
 
             # compute the joint probability distribution for the input space
@@ -958,13 +945,12 @@ def _fit(self, X, skip_num_points=0):
             # Free the memory used by the ball_tree
             del knn
 
-            if self.square_distances is True or self.metric == "euclidean":
-                # knn return the euclidean distance but we need it squared
-                # to be consistent with the 'exact' method. Note that the
-                # the method was derived using the euclidean method as in the
-                # input space. Not sure of the implication of using a different
-                # metric.
-                distances_nn.data **= 2
+            # knn return the euclidean distance but we need it squared
+            # to be consistent with the 'exact' method. Note that the
+            # the method was derived using the euclidean method as in the
+            # input space. Not sure of the implication of using a different
+            # metric.
+            distances_nn.data **= 2
 
             # compute the joint probability distribution for the input space
             P = _joint_probabilities_nn(distances_nn, self.perplexity, self.verbose)