def test_empty_arrays(self):
# Empty arrays of any type (including float) should be allowed
for dtype in self.dtypes_to_test:
target = np.array([], dtype=dtype)
converted = util.safe_np_int_cast([], dtype=dtype)
assert pickle.dumps(converted) == pickle.dumps(target)
target = np.array([[]], dtype=dtype)
converted = util.safe_np_int_cast([[]], dtype=dtype)
assert pickle.dumps(converted) == pickle.dumps(target)
def test_bad_types(self):
# Shouldn't be able to convert a float (possibility of rounding error)
for dtype in self.dtypes_to_test:
for bad_type in [
[0.1],
["str"],
{},
[{}],
np.array([0, 1], dtype=float),
]:
with pytest.raises(TypeError):
util.safe_np_int_cast(bad_type, dtype)
def test_nonrectangular_input(self):
bad_inputs = [
[0, 1, [2]],
[[0, 1, 2], []],
[(0, 1, 2), [2, 3]],
[(0, 1, 2), tuple()],
[(0, 1, 2), (2, )],
[(0, 1, 2), [2, 3]]]
for dtype in self.dtypes_to_test:
for bad_input in bad_inputs:
with self.assertRaises(TypeError):
util.safe_np_int_cast(bad_input, dtype)
def test_basic_arrays(self):
# Simple array
for dtype in self.dtypes_to_test:
target = np.array([0, 1], dtype=dtype)
for test_array in [[0, 1], (0, 1), np.array([0, 1]), target]:
converted = util.safe_np_int_cast(test_array, dtype=dtype)
# Use pickle to test exact equality including dtype
assert pickle.dumps(converted) == pickle.dumps(target)
# Nested array
target = np.array([[0, 1], [2, 3]], dtype=dtype)
for test_array in [[[0, 1], [2, 3]], np.array([[0, 1], [2, 3]]), target]:
converted = util.safe_np_int_cast(test_array, dtype=dtype)
assert pickle.dumps(converted) == pickle.dumps(target)
def test_copy(self):
# Check that a copy is not returned if copy=False & the original matches
# the specs
for dtype in self.dtypes_to_test:
for orig in (np.array([0, 1], dtype=dtype), np.array([], dtype=dtype)):
converted = util.safe_np_int_cast(orig, dtype=dtype, copy=True)
assert id(orig) != id(converted)
converted = util.safe_np_int_cast(orig, dtype=dtype, copy=False)
assert id(orig) == id(converted)
for dtype in [d for d in self.dtypes_to_test if d != np.int64]:
# non numpy arrays, or arrays of a different dtype don't get converted
for orig in ([0, 1], np.array([0, 1], dtype=np.int64)):
converted = util.safe_np_int_cast(orig, dtype=dtype, copy=False)
assert id(orig) != id(converted)
def test_nonrectangular_input(self):
bad_inputs = [
[0, 1, [2]],
[[0, 1, 2], []],
[(0, 1, 2), [2, 3]],
[(0, 1, 2), tuple()],
[(0, 1, 2), (2,)],
[(0, 1, 2), [2, 3]],
]
for dtype in self.dtypes_to_test:
for bad_input in bad_inputs:
# On some platforms and Python / numpy versions, a ValueError
# occurs instead
with self.assertRaises((TypeError, ValueError)):
util.safe_np_int_cast(bad_input, dtype)
def test_overflow(self):
for dtype in self.dtypes_to_test:
for bad_node in [np.iinfo(dtype).min - 1, np.iinfo(dtype).max + 1]:
self.assertRaises( # Test plain array
OverflowError, util.safe_np_int_cast, [0, bad_node], dtype)
self.assertRaises( # Test numpy array
OverflowError, util.safe_np_int_cast, np.array([0, bad_node]), dtype)
for good_node in [np.iinfo(dtype).min, np.iinfo(dtype).max]:
target = np.array([good_node], dtype=dtype)
self.assertEqual( # Test plain array
pickle.dumps(target),
pickle.dumps(util.safe_np_int_cast([good_node], dtype)))
self.assertEqual( # Test numpy array
pickle.dumps(target),
pickle.dumps(util.safe_np_int_cast(np.array([good_node]), dtype)))
def test_overflow(self):
for dtype in self.dtypes_to_test:
for bad_node in [np.iinfo(dtype).min - 1, np.iinfo(dtype).max + 1]:
with pytest.raises(OverflowError):
util.safe_np_int_cast([0, bad_node], dtype)
with pytest.raises(OverflowError):
util.safe_np_int_cast(np.array([0, bad_node]), dtype)
for good_node in [np.iinfo(dtype).min, np.iinfo(dtype).max]:
target = np.array([good_node], dtype=dtype)
assert pickle.dumps(target) == pickle.dumps(
util.safe_np_int_cast([good_node], dtype))
assert pickle.dumps(target) == pickle.dumps(
util.safe_np_int_cast(np.array([good_node]), dtype))
