Merge pull request scipy#7905 from apbard/REV-restore-wminkowski

REV: restore wminkowski

doc/release/1.0.0-notes.rst

@@ -187,6 +187,18 @@ The ``fillvalue`` of `scipy.signal.convolve2d` will be cast directly to the
 dtypes of the input arrays in the future and checked that it is a scalar or
 an array with a single element.
 
+``scipy.spatial.distance.matching`` is deprecated.  It is an alias of
+`scipy.spatial.distance.hamming`, which should be used instead.
+
+Implementation of `scipy.spatial.distance.wminkowski` was based on a wrong
+interpretation of the metric definition. In scipy 1.0 it has been just
+deprecated in the documentation to keep retro-compatibility but is recommended
+to use the new version of `scipy.spatial.distance.minkowski` that implements
+the correct behaviour.
+
+Positional arguments of `scipy.spatial.distance.pdist` and
+`scipy.spatial.distance.cdist` should be replaced with their keyword version.
+
 
 Backwards incompatible changes
 ==============================

scipy/spatial/distance.py

@@ -443,7 +443,7 @@ def minkowski(u, v, p=2, w=None):
        {||u-v||}_p = (\\sum{|u_i - v_i|^p})^{1/p}.
 
 
-       \\left(\\sum{(|w_i (u_i - v_i)|^p)}\\right)^{1/p}.
+       \\left(\\sum{w_i(|(u_i - v_i)|^p)}\\right)^{1/p}.
 
     Parameters
     ----------
@@ -486,18 +486,41 @@ def minkowski(u, v, p=2, w=None):
 # deprecated `wminkowski`.  Not done at once because it would be annoying for
 # downstream libraries that used `wminkowski` and support multiple scipy
 # versions.
-def wminkowski(*args, **kwds):
-    return minkowski(*args, **kwds)
+def wminkowski(u, v, p, w):
+    """
+    Computes the weighted Minkowski distance between two 1-D arrays.
+    The weighted Minkowski distance between `u` and `v`, defined as
+
+    .. math::
+
+       \\left(\\sum{(|w_i (u_i - v_i)|^p)}\\right)^{1/p}.
+
+    Parameters
+    ----------
+    u : (N,) array_like
+        Input array.
+    v : (N,) array_like
+        Input array.
+    p : int
+        The order of the norm of the difference :math:`{||u-v||}_p`.
+    w : (N,) array_like
+        The weight vector.
 
-if minkowski.__doc__ is not None:
-    doc = minkowski.__doc__.replace("Minkowski", "Weighted Minkowski")
-    doc += """Notes
+    Returns
+    -------
+    wminkowski : double
+        The weighted Minkowski distance between vectors `u` and `v`.
+
+    Notes
     -----
-    `wminkowski` is DEPRECATED.  It is simply an alias of `minkowski` in
-    scipy >= 1.0.
+    `wminkowski` is DEPRECATED. It implements a definition where weights
+    are powered. It is recommended to use the weighted version of `minkowski`
+    instead. This function will be removed in a future version of scipy.
 
     """
-    wminkowski.__doc__ = doc
+    w = _validate_vector(w)
+    return minkowski(u, v, p=p, w=w**p)
+
 
 def euclidean(u, v, w=None):
     """
@@ -509,7 +532,7 @@ def euclidean(u, v, w=None):
 
        {||u-v||}_2
 
-       \\left(\\sum{(|w_i (u_i - v_i)|^2)}\\right)^{1/2}
+       \\left(\\sum{(w_i |(u_i - v_i)|^2)}\\right)^{1/2}
 
     Parameters
     ----------
@@ -1675,8 +1698,6 @@ def pdist(X, metric='euclidean', *args, **kwargs):
             if(mstr in _METRICS['seuclidean'].aka or
                mstr in _METRICS['mahalanobis'].aka):
                 raise ValueError("metric %s incompatible with weights" % mstr)
-            if mstr in _METRICS['wminkowski'].aka:
-                mstr = "minkowski"
             # need to use python version for weighting
             kwargs['out'] = out
             mstr = "test_%s" % mstr
@@ -2404,8 +2425,6 @@ def cdist(XA, XB, metric='euclidean', *args, **kwargs):
             if(mstr in _METRICS['seuclidean'].aka or
                mstr in _METRICS['mahalanobis'].aka):
                 raise ValueError("metric %s incompatible with weights" % mstr)
-            if mstr in _METRICS['wminkowski'].aka:
-                mstr = "minkowski"
             # need to use python version for weighting
             kwargs['out'] = out
             mstr = "test_%s" % mstr

scipy/spatial/tests/test_distance.py

@@ -58,7 +58,8 @@
                                     hamming, jaccard, kulsinski, mahalanobis,
                                     matching, minkowski, rogerstanimoto,
                                     russellrao, seuclidean, sokalmichener,
-                                    sokalsneath, sqeuclidean, yule, wminkowski)
+                                    sokalsneath, sqeuclidean, yule)
+from scipy.spatial.distance import wminkowski as old_wminkowski
 
 _filenames = [
               "cdist-X1.txt",
@@ -221,7 +222,7 @@ def _assert_within_tol(a, b, atol=0, rtol=0, verbose_=False):
 
 def _rand_split(arrays, weights, axis, split_per, seed=None):
     # inverse operation for stats.collapse_weights
-    weights = np.array(weights)  # modified inplace; need a copy
+    weights = np.array(weights, dtype=np.float64)  # modified inplace; need a copy
     seeded_rand = np.random.RandomState(seed)
 
     def mytake(a, ix, axis):
@@ -236,7 +237,7 @@ def mytake(a, ix, axis):
         prev_w = weights[split_ix]
         q = seeded_rand.rand()
         weights[split_ix] = q * prev_w
-        weights = np.append(weights, (1-q) * prev_w)
+        weights = np.append(weights, (1. - q) * prev_w)
         arrays = [np.append(a, mytake(a, split_ix, axis=axis),
                             axis=axis) for a in arrays]
     return arrays, weights
@@ -362,7 +363,7 @@ def wrapped(*args, **kwargs):
 wdice = _weight_checked(dice)
 wcosine = _weight_checked(cosine)
 wcorrelation = _weight_checked(correlation)
-wminkowski = _weight_checked(wminkowski, const_test=False)
+wminkowski = _weight_checked(minkowski, const_test=False)
 wjaccard = _weight_checked(jaccard)
 weuclidean = _weight_checked(euclidean, const_test=False)
 wsqeuclidean = _weight_checked(sqeuclidean, const_test=False)
@@ -533,6 +534,8 @@ def test_cdist_calling_conventions(self):
             for metric in _METRICS_NAMES:
                 if verbose > 2:
                     print("testing: ", metric, " with: ", eo_name)
+                if metric == 'wminkowski':
+                    continue
                 if metric in {'dice', 'yule', 'kulsinski', 'matching',
                               'rogerstanimoto', 'russellrao', 'sokalmichener',
                               'sokalsneath'} and 'bool' not in eo_name:
@@ -1294,6 +1297,8 @@ def test_pdist_calling_conventions(self):
             # NOTE: num samples needs to be > than dimensions for mahalanobis
             X = eo[eo_name][::5, ::2]
             for metric in _METRICS_NAMES:
+                if metric == 'wminkowski':
+                    continue
                 if verbose > 2:
                     print("testing: ", metric, " with: ", eo_name)
                 if metric in {'dice', 'yule', 'kulsinski', 'matching',
@@ -1391,6 +1396,24 @@ def test_minkowski(self):
                 assert_almost_equal(dist1p5, (1.0 + 2.0**1.5)**(2. / 3))
                 dist2 = wminkowski(x, y, p=2)
 
+    def test_old_wminkowski(self):
+        with suppress_warnings() as wrn:
+            wrn.filter(message="`wminkowski` is deprecated")
+            w = np.array([1.0, 2.0, 0.5])
+            for x, y in self.cases:
+                dist1 = old_wminkowski(x, y, p=1, w=w)
+                assert_almost_equal(dist1, 3.0)
+                dist1p5 = old_wminkowski(x, y, p=1.5, w=w)
+                assert_almost_equal(dist1p5, (2.0**1.5+1.0)**(2./3))
+                dist2 = old_wminkowski(x, y, p=2, w=w)
+                assert_almost_equal(dist2, np.sqrt(5))
+
+            # test weights Issue #7893
+            arr = np.arange(4)
+            w = np.full_like(arr, 4)
+            assert_almost_equal(old_wminkowski(arr, arr + 1, p=2, w=w), 8.0)
+            assert_almost_equal(wminkowski(arr, arr + 1, p=2, w=w), 4.0)
+
     def test_euclidean(self):
         for x, y in self.cases:
             dist = weuclidean(x, y)