Use np.histogram_bin_edges with NumPy >= 1.15.0

Before that function was exposed, finding the bin edges required an
unnecessary calculation of the combined histogram. We now detect the
NumPy version and use `np.histogram_bin_edges()` if it's available.

Author: wmayner

README.rst

@@ -172,11 +172,12 @@ Limitations and Caveats
 
 - ``emd_samples()``:
 
-  - Using the default ``bins='auto'`` results in an extra call to
-    ``np.histogram()`` to determine the bin lengths, since `the NumPy
-    bin-selectors are not exposed in the public API
+  - With ``numpy < 1.15.0``, using the default ``bins='auto'`` results in an
+    extra call to ``np.histogram()`` to determine the bin lengths, since `the
+    NumPy bin-selectors are not exposed in the public API
     <https://github.com/numpy/numpy/issues/10183>`_. For performance, you may
-    want to set the bins yourself.
+    want to set the bins yourself. If ``numpy >= 1.15`` is available,
+    ``np.histogram_bin_edges()`` is called instead, which is more efficient.
 
 
 Contributing

pyemd/emd.pyx

@@ -3,6 +3,8 @@
 # distutils: language = c++
 # emd.pyx
 
+from pkg_resources import parse_version
+
 from libcpp.pair cimport pair
 from libcpp.vector cimport vector
 import cython
@@ -139,6 +141,16 @@ def euclidean_pairwise_distance_matrix(x):
     return distance_matrix.reshape(len(x), len(x))
 
 
+# Use `np.histogram_bin_edges` if available (since NumPy version 1.15.0)
+if parse_version(np.__version__) >= parse_version('1.15.0'):
+    get_bins = np.histogram_bin_edges
+else:
+    def get_bins(a, bins=10, **kwargs):
+        if isinstance(bins, str):
+            hist, bins = np.histogram(a, bins=bins, **kwargs)
+        return bins
+
+
 def emd_samples(first_array,
                 second_array,
                 extra_mass_penalty=DEFAULT_EXTRA_MASS_PENALTY,
@@ -196,14 +208,10 @@ def emd_samples(first_array,
     if range is None:
         range = (min(np.min(first_array), np.min(second_array)),
                  max(np.max(first_array), np.max(second_array)))
-    # Use automatic binning from `np.histogram()`
-    # TODO: Use `np.histogram_bin_edges()` when it's available;
-    # see https://github.com/numpy/numpy/issues/10183
-    if isinstance(bins, str):
-        hist, _ = np.histogram(np.concatenate([first_array, second_array]),
-                               range=range,
-                               bins=bins)
-        bins = len(hist)
+    # Get bin edges using both arrays
+    bins = get_bins(np.concatenate([first_array, second_array]),
+                    range=range,
+                    bins=bins)
     # Compute histograms
     first_histogram, bin_edges = np.histogram(first_array,
                                               range=range,

tox.ini

@@ -1,7 +1,12 @@
 [tox]
-envlist = py{27,34,35,36}
+envlist = py{27,34,35,36}-numpy{114,115}
 
 [testenv]
-deps = -r{toxinidir}/test_requirements.txt
+deps =
+  -r{toxinidir}/test_requirements.txt
+  # Use NumPy < 1.14 and NumPy >= 1.15 for `get_bins()` switch
+  numpy114: numpy<1.15
+  numpy115: numpy>=1.15
+
 commands = make test
 whitelist_externals = make