xhluca
diff --git a/‎doc/whats_new/v0.21.rst
Lines changed: 0 additions & 9 deletions b/‎doc/whats_new/v0.21.rst
Lines changed: 0 additions & 9 deletions
diff --git a/‎sklearn/manifold/t_sne.py
Lines changed: 7 additions & 16 deletions b/‎sklearn/manifold/t_sne.py
Lines changed: 7 additions & 16 deletions
@@ -162,15 +162,6 @@ Support for Python 3.4 and below has been officially dropped.
 - |Fix| Fixed a bug in the 'saga' solver where the weights would not be
   correctly updated in some cases. :issue:`11646` by `Tom Dupre la Tour`_.
 
-:mod:`sklearn.manifold`
-............................
-
-- |Efficiency| Make :func:`tsne.trustworthiness` use an inverted index
-  instead of an `np.where` lookup to find the rank of neighbors in the input
-  space. This improves trustworthiness in particular when computed with
-  lots of neighbors and/or small datasets.
-  :issue:`9907` by :user:`William de Vazelhes <wdevazelhes>`.
-
 Multiple modules
 ................
 
 
@@ -454,27 +454,18 @@ def trustworthiness(X, X_embedded, n_neighbors=5,
                       "parameter instead.", DeprecationWarning)
         metric = 'precomputed'
     dist_X = pairwise_distances(X, metric=metric)
-    if metric == 'precomputed':
-        dist_X = dist_X.copy()
-    # we set the diagonal to np.inf to exclude the points themselves from
-    # their own neighborhood
-    np.fill_diagonal(dist_X, np.inf)
     ind_X = np.argsort(dist_X, axis=1)
-    # `ind_X[i]` is the index of sorted distances between i and other samples
     ind_X_embedded = NearestNeighbors(n_neighbors).fit(X_embedded).kneighbors(
         return_distance=False)
 
-    # We build an inverted index of neighbors in the input space: For sample i,
-    # we define `inverted_index[i]` as the inverted index of sorted distances:
-    # inverted_index[i][ind_X[i]] = np.arange(1, n_sample + 1)
     n_samples = X.shape[0]
-    inverted_index = np.zeros((n_samples, n_samples), dtype=int)
-    ordered_indices = np.arange(n_samples + 1)
-    inverted_index[ordered_indices[:-1, np.newaxis],
-                   ind_X] = ordered_indices[1:]
-    ranks = inverted_index[ordered_indices[:-1, np.newaxis],
-                           ind_X_embedded] - n_neighbors
-    t = np.sum(ranks[ranks > 0])
+    t = 0.0
+    ranks = np.zeros(n_neighbors)
+    for i in range(n_samples):
+        for j in range(n_neighbors):
+            ranks[j] = np.where(ind_X[i] == ind_X_embedded[i, j])[0][0]
+        ranks -= n_neighbors
+        t += np.sum(ranks[ranks > 0])
     t = 1.0 - t * (2.0 / (n_samples * n_neighbors *
                           (2.0 * n_samples - 3.0 * n_neighbors - 1.0)))
     return t