Set TSNE's internal PCA to always use numpy as output

betatim · betatim · commit 5f2cb8166e31 · 2023-01-12T15:14:23.000+01:00
The internal PCA should always output a numpy array as it is results
are only used internally.
diff --git a/doc/whats_new/v1.2.rst b/doc/whats_new/v1.2.rst
@@ -30,6 +30,13 @@ Changelog
   `verbose` parameter set to a value greater than 0.
   :pr:`25250` by :user:`Jérémie Du Boisberranger <jeremiedbb>`.
 
+:mod:`sklearn.manifold`
+...........................
+
+- |Fix| :class:`manifold.TSNE` now works correctly when output type is
+  set to pandas :pr:`25370` by :user:`Tim Head <betatim>`.
+
+
 :mod:`sklearn.preprocessing`
 ............................
 
diff --git a/sklearn/manifold/_t_sne.py b/sklearn/manifold/_t_sne.py
@@ -990,6 +990,8 @@ def _fit(self, X, skip_num_points=0):
                 svd_solver="randomized",
                 random_state=random_state,
             )
+            # Alays output a numpy array, no matter what is configured globally
+            pca.set_output(transform="default")
             X_embedded = pca.fit_transform(X).astype(np.float32, copy=False)
             # PCA is rescaled so that PC1 has standard deviation 1e-4 which is
             # the default value for random initialization. See issue #18018.
diff --git a/sklearn/manifold/tests/test_t_sne.py b/sklearn/manifold/tests/test_t_sne.py
@@ -5,6 +5,7 @@
 import scipy.sparse as sp
 import pytest
 
+from sklearn import config_context
 from sklearn.neighbors import NearestNeighbors
 from sklearn.neighbors import kneighbors_graph
 from sklearn.exceptions import EfficiencyWarning
@@ -1191,3 +1192,13 @@ def test_tsne_perplexity_validation(perplexity):
     msg = "perplexity must be less than n_samples"
     with pytest.raises(ValueError, match=msg):
         est.fit_transform(X)
+
+
+def test_tsne_works_with_pandas_output():
+    """Make sure that TSNE works when the output is set to "pandas"
+
+    Non-regression test for gh-25365.
+    """
+    with config_context(transform_output="pandas"):
+        arr = np.arange(35 * 4).reshape(35, 4)
+        TSNE(n_components=2).fit_transform(arr)

Original file line number	Diff line number	Diff line change
`@@ -990,6 +990,8 @@ def _fit(self, X, skip_num_points=0):`
`990`	`990`	`svd_solver="randomized",`
`991`	`991`	`random_state=random_state,`
`992`	`992`	`)`
	`993`	`+ # Alays output a numpy array, no matter what is configured globally`
	`994`	`+ pca.set_output(transform="default")`
`993`	`995`	`X_embedded = pca.fit_transform(X).astype(np.float32, copy=False)`
`994`	`996`	`# PCA is rescaled so that PC1 has standard deviation 1e-4 which is`
`995`	`997`	`# the default value for random initialization. See issue #18018.`