ENH: When histogramming data with integer dtype, force bin width >= 1.

anntzer · anntzer · commit e53f68337c07 · 2019-01-05T12:23:04.000+01:00
Bins of width &lt; 1 don't make sense for integer data, they just add a
bunch of spurious, unpopulated bins.  (Perhaps an even better
improvement would be to make sure that, when using integer data, the
binwidth is also integer, so that each bin always covers the same number
of possible values, but I guess that's possibly a more domain-specific
issue.)

Before the PR:

    In [1]: np.histogram_bin_edges(np.tile(np.arange(10), 1000), "auto")
    Out[1]:
    array([0.  , 0.45, 0.9 , 1.35, 1.8 , 2.25, 2.7 , 3.15, 3.6 , 4.05, 4.5 ,
           4.95, 5.4 , 5.85, 6.3 , 6.75, 7.2 , 7.65, 8.1 , 8.55, 9.  ])

After:

    In [1]: np.histogram_bin_edges(np.tile(np.arange(10), 1000), "auto")
    Out[1]: array([0., 1., 2., 3., 4., 5., 6., 7., 8., 9.])
diff --git a/numpy/lib/histograms.py b/numpy/lib/histograms.py
@@ -346,6 +346,8 @@ def _get_bin_edges(a, bins, range, weights):
             # Do not call selectors on empty arrays
             width = _hist_bin_selectors[bin_name](a)
             if width:
+                if np.issubdtype(a.dtype, np.integer) and width < 1:
+                    width = 1
                 n_equal_bins = int(np.ceil(_unsigned_subtract(last_edge, first_edge) / width))
             else:
                 # Width can be zero for some estimators, e.g. FD when
@@ -553,6 +555,9 @@ def histogram_bin_edges(a, bins=10, range=None, weights=None):
         The simplest and fastest estimator. Only takes into account the
         data size.
 
+    Additionally, if the data is of integer dtype, then the binwidth will never
+    be less than 1.
+
     Examples
     --------
     >>> arr = np.array([0, 0, 0, 1, 2, 3, 3, 4, 5])
diff --git a/numpy/lib/tests/test_histograms.py b/numpy/lib/tests/test_histograms.py
@@ -478,7 +478,7 @@ def test_small(self):
                          'doane': 3, 'sqrt': 2}}
 
         for testlen, expectedResults in small_dat.items():
-            testdat = np.arange(testlen)
+            testdat = np.arange(testlen).astype(float)
             for estimator, expbins in expectedResults.items():
                 a, b = np.histogram(testdat, estimator)
                 assert_equal(len(a), expbins, err_msg="For the {0} estimator "
@@ -575,6 +575,27 @@ def test_simple_range(self):
                 msg += " with datasize of {0}".format(testlen)
                 assert_equal(len(a), numbins, err_msg=msg)
 
+    def test_integer(self):
+        """
+        Test that bin width for integer data is at least 1.
+        """
+        estimator_list = ['fd', 'scott', 'rice', 'sturges', 'auto']
+        for estimator in estimator_list:
+            assert_equal(
+                np.histogram_bin_edges(np.tile(np.arange(9), 1000), estimator),
+                np.arange(9))
+
+    def test_integer_non_auto(self):
+        """
+        Test that the bin-width>=1 requirement *only* applies to auto binning.
+        """
+        assert_equal(
+            np.histogram_bin_edges(np.tile(np.arange(9), 1000), 16),
+            np.arange(17) / 2)
+        assert_equal(
+            np.histogram_bin_edges(np.tile(np.arange(9), 1000), [.1, .2]),
+            [.1, .2])
+
     def test_simple_weighted(self):
         """
         Check that weighted data raises a TypeError
