minor update

TomDLT · TomDLT · commit e75ea2837f16 · 2018-01-26T18:52:08.000+01:00
diff --git a/doc/modules/neighbors.rst b/doc/modules/neighbors.rst
@@ -544,9 +544,9 @@ include the precomputation in a pipeline:
     >>> from sklearn.neighbors import KNeighborsTransformer
     >>> from sklearn.pipeline import make_pipeline
     >>> estimator = make_pipeline(
-            KNeighborsTransformer(n_neighbors=5, mode='distance'),
-            Isomap(neighbors_algorithm='precomputed'),
-            memory='/path/to/cache')
+    ...     KNeighborsTransformer(n_neighbors=5, mode='distance'),
+    ...     Isomap(neighbors_algorithm='precomputed'),
+    ...     memory='/path/to/cache')
 
 Second, these transformers give finer control on the nearest neighbors
 estimation, for instance enabling multiprocessing though the parameter `n_jobs`.
diff --git a/examples/neighbors/neighbors_in_pipeline_api.py b/examples/neighbors/neighbors_in_pipeline_api.py
@@ -144,7 +144,7 @@ def run_benchmark():
     n_iter = 250
     perplexity = 30
     # TSNE requires a certain number of neighbors which depends on the
-    # perplexity parameter
+    # perplexity parameter
     n_neighbors = int(3. * perplexity + 1)
 
     transformers = {
diff --git a/sklearn/manifold/tests/test_t_sne.py b/sklearn/manifold/tests/test_t_sne.py
@@ -172,7 +172,7 @@ def test_pipeline_with_nearest_neighbors_transformer():
     X = rng.randn(20, 2)
 
     # XXX: This test currently fail for euclidean metric since it's not squared
-    for metric in ['minkowski']:
+    for metric in ['minkowski', 'sqeuclidean']:
 
         # compare the chained version and the compact version
         est_chain = make_pipeline(
diff --git a/sklearn/neighbors/base.py b/sklearn/neighbors/base.py
@@ -196,7 +196,7 @@ def _kneighbors_from_graph(graph, n_neighbors):
             this_dist = graph.data[idx]
             this_ind = graph.indices[idx]
             # sort neigh_dist and neigh_ind to have increasing distances
-            perm = np.argsort(neigh_dist)
+            perm = np.argsort(this_dist)
             neigh_dist[i] = this_dist[perm][:n_neighbors]
             neigh_ind[i] = this_ind[perm][:n_neighbors]
         return neigh_dist, neigh_ind
diff --git a/sklearn/neighbors/classification.py b/sklearn/neighbors/classification.py
@@ -10,7 +10,6 @@
 
 import numpy as np
 from scipy import stats
-import scipy.sparse as sp
 from ..utils.extmath import weighted_mode
 
 from .base import \
diff --git a/sklearn/neighbors/tests/test_neighbors.py b/sklearn/neighbors/tests/test_neighbors.py
@@ -201,16 +201,15 @@ def test_precomputed_sparse_invalid():
     dist = np.array([[0., 2., 0.], [2., 0., 3.], [0., 3., 0.]])
     dist_csr = csr_matrix(dist)
     neigh.fit(dist_csr)
-    assert_raises_regex(ValueError, "Not enough neighbors in"
-                        " .* to get 2 nearest neighbors.*", neigh.kneighbors,
-                        None, n_neighbors=2)
+    msg = "2 neighbors per samples are required, but some samples have only 1"
+    assert_raises_regex(ValueError, msg, neigh.kneighbors, None, n_neighbors=2)
 
     # Checks error with inconsistent distance matrix
     dist = np.array([[5., 2., 1.], [-2., 0., 3.], [1., 3., 0.]])
     dist_csr = csr_matrix(dist)
-    assert_raises_regex(ValueError, "Not a valid distance"
-                        " .*non-negative values.*", neigh.kneighbors,
-                        dist_csr, n_neighbors=1)
+    msg = "Not a valid distance .*non-negative values.*"
+    assert_raises_regex(ValueError, msg, neigh.kneighbors, dist_csr,
+                        n_neighbors=1)
 
 
 def test_precomputed_cross_validation():
