ENH: Make assert_almost_equal & assert_array_almost_equal consistent.

This changes the check for scalars in assert_almost_equal so that

    abs(actual - desired) < 1.5 * 10**(-decimal)

Note that the previous documentation claimed that the functions were
equivalent to

    abs(actual - desired) < .5 * 10**(-decimal)

but that was not how they behaved in practice.

Due to the change in implementation, some very delicate tests may fail
that did not fail before. No extra failures were noted in scipy.

Closes #5200.

1 files changed, 19 insertions, 14 deletions

diff --git a/numpy/testing/utils.py b/numpy/testing/utils.py
index 3f4a8568a..dfed5d148 100644
--- a/numpy/testing/utils.py
+++ b/numpy/testing/utils.py
@@ -424,11 +424,14 @@ def assert_almost_equal(actual,desired,decimal=7,err_msg='',verbose=True):
               instead of this function for more consistent floating point
               comparisons.
 
-    The test is equivalent to ``abs(desired-actual) < 0.5 * 10**(-decimal)``.
+    The test verifies that the elements of ``actual`` and ``desired`` satisfy.
 
-    Given two objects (numbers or ndarrays), check that all elements of these
-    objects are almost equal. An exception is raised at conflicting values.
-    For ndarrays this delegates to assert_array_almost_equal
+        ``abs(desired-actual) < 1.5 * 10**(-decimal)``
+
+    That is a looser test than originally documented, but agrees with what the
+    actual implementation in `assert_array_almost_equal` did up to rounding
+    vagaries. An exception is raised at conflicting values. For ndarrays this
+    delegates to assert_array_almost_equal
 
     Parameters
     ----------
@@ -529,7 +532,7 @@ def assert_almost_equal(actual,desired,decimal=7,err_msg='',verbose=True):
             return
     except (NotImplementedError, TypeError):
         pass
-    if round(abs(desired - actual), decimal) != 0:
+    if abs(desired - actual) >= 1.5 * 10.0**(-decimal):
         raise AssertionError(_build_err_msg())
 
 
@@ -819,14 +822,16 @@ def assert_array_almost_equal(x, y, decimal=6, err_msg='', verbose=True):
               instead of this function for more consistent floating point
               comparisons.
 
-    The test verifies identical shapes and verifies values with
-    ``abs(desired-actual) < 0.5 * 10**(-decimal)``.
+    The test verifies identical shapes and that the elements of ``actual`` and
+    ``desired`` satisfy.
 
-    Given two array_like objects, check that the shape is equal and all
-    elements of these objects are almost equal. An exception is raised at
-    shape mismatch or conflicting values. In contrast to the standard usage
-    in numpy, NaNs are compared like numbers, no assertion is raised if
-    both objects have NaNs in the same positions.
+        ``abs(desired-actual) < 1.5 * 10**(-decimal)``
+
+    That is a looser test than originally documented, but agrees with what the
+    actual implementation did up to rounding vagaries. An exception is raised
+    at shape mismatch or conflicting values. In contrast to the standard usage
+    in numpy, NaNs are compared like numbers, no assertion is raised if both
+    objects have NaNs in the same positions.
 
     Parameters
     ----------
@@ -903,12 +908,12 @@ def assert_array_almost_equal(x, y, decimal=6, err_msg='', verbose=True):
         # casting of x later.
         dtype = result_type(y, 1.)
         y = array(y, dtype=dtype, copy=False, subok=True)
-        z = abs(x-y)
+        z = abs(x - y)
 
         if not issubdtype(z.dtype, number):
             z = z.astype(float_)  # handle object arrays
 
-        return around(z, decimal) <= 10.0**(-decimal)
+        return z < 1.5 * 10.0**(-decimal)
 
     assert_array_compare(compare, x, y, err_msg=err_msg, verbose=verbose,
              header=('Arrays are not almost equal to %d decimals' % decimal),