Further tests cleanup; fix xfail wrapper

Author: ml31415

numpy_groupies/aggregate_pandas.py

@@ -39,6 +39,8 @@ def _wrapper(group_idx, a, size, fill_value, func='sum', dtype=None, ddof=0, **k
                   nanlen=partial(_wrapper, func='count'),
                   argmax=partial(_wrapper, func='idxmax'),
                   argmin=partial(_wrapper, func='idxmin'),
+                  nanargmax=partial(_wrapper, func='idxmax'),
+                  nanargmin=partial(_wrapper, func='idxmin'),
                   generic=_wrapper)
 
 

numpy_groupies/tests/__init__.py

@@ -26,33 +26,37 @@ def _impl_name(impl):
 
 
 _not_implemented_by_impl_name = {
-    'numpy': ['cumprod','cummax', 'cummin'],
-    'purepy': ['cumprod','cummax', 'cummin'],
-    'numba': ('array', 'list'),
-    'pandas': ('array', 'list'),
-    'weave':  ('argmin', 'argmax', 'array', 'list', 'cumsum',
-               '<lambda>', 'func_preserve_order', 'func_arbitrary')}
+    'numpy':  ('cumprod','cummax', 'cummin'),
+    'purepy': ('cumsum', 'cumprod','cummax', 'cummin', 'sumofsquares'),
+    'numba':  ('array', 'list', 'sort'),
+    'pandas': ('array', 'list', 'sort', 'sumofsquares', 'nansumofsquares'),
+    'weave':  ('argmin', 'argmax', 'array', 'list', 'sort', 'cumsum', 'cummax', 'cummin',
+               'nanargmin', 'nanargmax', 'sumofsquares', 'nansumofsquares',
+               '<lambda>', 'custom_callable'),
+    'ufunc':  'NO_CHECK'}
+
 
 def _wrap_notimplemented_xfail(impl, name=None):
     """Some implementations lack some functionality. That's ok, let's xfail that instead of raising errors."""
 
-    def _try_xfail(*args, **kwargs):
+    def try_xfail(*args, **kwargs):
         try:
             return impl(*args, **kwargs)
         except NotImplementedError as e:
+            impl_name = impl.__module__.split('_')[-1]
             func = kwargs.pop('func', None)
             if callable(func):
                 func = func.__name__
-            wrap_funcs = _not_implemented_by_impl_name.get(func, None)
-            if wrap_funcs is None or func in wrap_funcs:
+            not_implemented_ok = _not_implemented_by_impl_name.get(impl_name, [])
+            if not_implemented_ok == 'NO_CHECK' or func in not_implemented_ok:
                 pytest.xfail("Functionality not implemented")
             else:
                 raise e
     if name:
-        _try_xfail.__name__ = name
+        try_xfail.__name__ = name
     else:
-        _try_xfail.__name__ = impl.__name__
-    return _try_xfail
+        try_xfail.__name__ = impl.__name__
+    return try_xfail
 
 
 func_list = ('sum', 'prod', 'min', 'max', 'all', 'any', 'mean', 'std', 'var', 'len',

numpy_groupies/tests/test_compare.py

@@ -67,6 +67,18 @@ def aggregate_cmp(request, seed=100):
     return AttrDict(locals())
 
 
+def _deselect_purepy(aggregate_cmp, *args, **kwargs):
+    # purepy implementation does not handle ndim arrays
+    # This is a won't fix and should be deselected instead of skipped
+    return aggregate_cmp.endswith('py')
+
+
+def _deselect_purepy_nanfuncs(aggregate_cmp, func, *args, **kwargs):
+    # purepy implementation does not handle nan values correctly
+    # This is a won't fix and should be deselected instead of skipped
+    return 'nan' in getattr(func, '__name__', func) and aggregate_cmp.endswith('py')
+
+
 def func_arbitrary(iterator):
     tmp = 0
     for x in iterator:
@@ -81,14 +93,10 @@ def func_preserve_order(iterator):
     return tmp
 
 
-def _deselect_purepy_nanfuncs(aggregate_cmp, func, fill_value):
-    # purepy implementation does not handle nan values correctly
-    return 'nan' in getattr(func, '__name__', func) and aggregate_cmp.endswith('py')
-
-
+@pytest.mark.filterwarnings("ignore:numpy.ufunc size changed")
 @pytest.mark.deselect_if(func=_deselect_purepy_nanfuncs)
-@pytest.mark.parametrize(["func", "fill_value"], product(func_list, [0, 1, np.nan]),
-                         ids=lambda x: getattr(x, '__name__', x))
+@pytest.mark.parametrize("fill_value", [0, 1, np.nan])
+@pytest.mark.parametrize("func", func_list, ids=lambda x: getattr(x, '__name__', x))
 def test_cmp(aggregate_cmp, func, fill_value, decimal=10):
     is_nanfunc = 'nan' in getattr(func, '__name__', func)
     a = aggregate_cmp.nana if is_nanfunc else aggregate_cmp.a
@@ -107,9 +115,16 @@ def test_cmp(aggregate_cmp, func, fill_value, decimal=10):
                 raise
         if isinstance(ref, np.ndarray):
             assert res.dtype == ref.dtype
-        np.testing.assert_allclose(res, ref, rtol=10**-decimal)
+        try:
+            np.testing.assert_allclose(res, ref, rtol=10**-decimal)
+        except AssertionError:
+            if 'arg' in func and aggregate_cmp.test_pair.startswith('pandas'):
+                pytest.xfail("pandas doesn't fill indices for all-nan groups with fill_value, but with -inf instead")
+            else:
+                raise
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 @pytest.mark.parametrize(["ndim", "order"], product([2, 3], ["C", "F"]))
 def test_cmp_ndim(aggregate_cmp, ndim, order, outsize=100, decimal=14):
     nindices = int(outsize ** ndim)

numpy_groupies/tests/test_generic.py

@@ -17,6 +17,20 @@ def aggregate_all(request):
     return _wrap_notimplemented_xfail(impl.aggregate, 'aggregate_' + name)
 
 
+def _deselect_purepy(aggregate_all, *args, **kwargs):
+    # purepy implementations does not handle nan values and ndim correctly.
+    # So it needs to be excluded from several tests."""
+    return aggregate_all.__name__.endswith('purepy')
+
+
+def _deselect_purepy_and_invalid_axis(aggregate_all, size, axis, *args, **kwargs):
+    if axis >= len(size):
+        return True
+    if aggregate_all.__name__.endswith('purepy'):
+        # purepy does not handle axis parameter
+        return True
+
+
 def test_preserve_missing(aggregate_all):
     res = aggregate_all(np.array([0, 1, 3, 1, 3]), np.arange(101, 106, dtype=int))
     np.testing.assert_array_equal(res, np.array([101, 206, 0, 208]))
@@ -100,12 +114,14 @@ def test_array_ordering(aggregate_all, order, size=10):
     assert aggregate_all(np.zeros(size, dtype=int), mat[0, :], order=order)[0] == sum(range(size))
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 @pytest.mark.parametrize("size", [None, (10, 2)])
 def test_ndim_group_idx(aggregate_all, size):
     group_idx = np.vstack((np.repeat(np.arange(10), 10), np.repeat([0, 1], 50)))
     aggregate_all(group_idx, 1, size=size)
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 @pytest.mark.parametrize(["ndim", "order"], itertools.product([1, 2, 3], ["C", "F"]))
 def test_ndim_indexing(aggregate_all, ndim, order, outsize=10):
     nindices = int(outsize ** ndim)
@@ -242,9 +258,8 @@ def test_argmin_argmax_nonans(aggregate_all):
     np.testing.assert_array_equal(res, [3, -1, -1, 5])
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 def test_argmin_argmax_nans(aggregate_all):
-    if aggregate_all.__name__.endswith('purepy'):
-        pytest.xfail("purepy doesn't handle nan values correctly")
     if aggregate_all.__name__.endswith('pandas'):
         pytest.xfail("pandas always ignores nans")
 
@@ -258,9 +273,10 @@ def test_argmin_argmax_nans(aggregate_all):
     np.testing.assert_array_equal(res, [3, -1, -1, -1])
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 def test_nanargmin_nanargmax_nans(aggregate_all):
-    if aggregate_all.__name__.endswith('purepy'):
-        pytest.xfail("purepy doesn't handle nan values correctly")
+    if aggregate_all.__name__.endswith('pandas'):
+        pytest.xfail("pandas doesn't fill indices for all-nan groups with fill_value but with -inf instead")
 
     group_idx = np.array([0, 0, 0, 0, 3, 3, 3, 3])
     a = np.array([4, 4, np.nan, 1, np.nan, np.nan, np.nan, np.nan])
@@ -339,12 +355,8 @@ def test_list_ordering(aggregate_all, order):
         a = a[::-1]
     ref = a[:4]
 
-    try:
-        res = aggregate_all(group_idx, a, func=list)
-    except NotImplementedError:
-        pytest.xfail("Function not yet implemented")
-    else:
-        np.testing.assert_array_equal(np.array(res[0]), ref)
+    res = aggregate_all(group_idx, a, func=list)
+    np.testing.assert_array_equal(np.array(res[0]), ref)
 
 
 @pytest.mark.parametrize("order", ["normal", "reverse"])
@@ -360,14 +372,6 @@ def test_sort(aggregate_all, order):
     np.testing.assert_array_equal(res, ref)
 
 
-def _deselect_purepy_and_invalid_axis(aggregate_all, func, size, axis):
-    if axis >= len(size):
-        return True
-    if aggregate_all.__name__.endswith('purepy'):
-        # purepy does not handle axis parameter
-        return True
-
-
 @pytest.mark.deselect_if(func=_deselect_purepy_and_invalid_axis)
 @pytest.mark.parametrize("axis", (0, 1))
 @pytest.mark.parametrize("size", ((12,), (12, 5)))
@@ -424,6 +428,7 @@ def test_along_axis(aggregate_all, func, size, axis):
     np.testing.assert_allclose(actual.squeeze(), expected)
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 def test_not_last_axis_reduction(aggregate_all):
     group_idx = np.array([1, 2, 2, 0, 1])
     a = np.array([
@@ -442,8 +447,9 @@ def test_not_last_axis_reduction(aggregate_all):
     np.testing.assert_allclose(expected, actual)
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 def test_custom_callable(aggregate_all):
-    def sum_(x):
+    def custom_callable(x):
         return x.sum()
 
     size = (10,)
@@ -453,12 +459,13 @@ def sum_(x):
     a = np.random.randn(*size)
 
     expected = a.sum(axis=axis, keepdims=True)
-    actual = aggregate_all(group_idx, a, axis=axis, func=sum_, fill_value=0)
+    actual = aggregate_all(group_idx, a, axis=axis, func=custom_callable, fill_value=0)
     assert actual.ndim == a.ndim
 
     np.testing.assert_allclose(actual, expected)
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 def test_argreduction_nD_array_1D_idx(aggregate_all):
     # https://github.com/ml31415/numpy-groupies/issues/41
     group_idx = np.array([0, 0, 2, 2, 2, 1, 1, 2, 2, 1, 1, 0], dtype=int)
@@ -468,6 +475,7 @@ def test_argreduction_nD_array_1D_idx(aggregate_all):
     np.testing.assert_equal(actual, expected)
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 def test_argreduction_negative_fill_value(aggregate_all):
     if aggregate_all.__name__.endswith('pandas'):
         pytest.xfail("pandas always skips nan values")
@@ -479,6 +487,7 @@ def test_argreduction_negative_fill_value(aggregate_all):
     np.testing.assert_equal(actual, expected)
 
 
+@pytest.mark.deselect_if(func=_deselect_purepy)
 @pytest.mark.parametrize("nan_inds", (None, tuple([[1, 4, 5], Ellipsis]), tuple((1, (0, 1, 2, 3)))))
 @pytest.mark.parametrize("ddof", (0, 1))
 @pytest.mark.parametrize("func", ("nanvar", "nanstd"))