Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Removed run_all_in_graph_and_eager_mode in sparsemax. #1689

Merged
merged 3 commits into from
Apr 17, 2020
Merged

Removed run_all_in_graph_and_eager_mode in sparsemax. #1689

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9f8107b
Removed run_all_in_graph_and_eager_mode in sparsemax.
gabrieldemarmiesse Apr 17, 2020
3daa416
Removed old code.
gabrieldemarmiesse Apr 17, 2020
1b4cd70
Removed useless astype.
gabrieldemarmiesse Apr 17, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
196 changes: 101 additions & 95 deletions tensorflow_addons/activations/tests/sparsemax_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,140 +79,146 @@ def test_sparsemax_against_numpy_low_rank(dtype):
assert np_sparsemax.shape == tf_sparsemax_out.shape


@test_utils.run_all_with_types(["float32", "float64"])
@test_utils.run_all_in_graph_and_eager_modes
class SparsemaxTest(tf.test.TestCase):
def _tf_sparsemax(self, z, dtype, **kwargs):
tf_sparsemax_op = sparsemax(z.astype(dtype), **kwargs)
tf_sparsemax_out = self.evaluate(tf_sparsemax_op)

return tf_sparsemax_op, tf_sparsemax_out
@pytest.mark.parametrize("dtype", ["float32", "float64"])
def test_sparsemax_against_numpy(dtype):
"""check sparsemax kernel against numpy."""
random = np.random.RandomState(1)

def test_sparsemax_against_numpy(self, dtype=None):
"""check sparsemax kernel against numpy."""
random = np.random.RandomState(1)
z = random.uniform(low=-3, high=3, size=(test_obs, 10))

z = random.uniform(low=-3, high=3, size=(test_obs, 10))
tf_sparsemax_out = sparsemax(z.astype(dtype))
np_sparsemax = _np_sparsemax(z).astype(dtype)

tf_sparsemax_op, tf_sparsemax_out = self._tf_sparsemax(z, dtype)
np_sparsemax = _np_sparsemax(z).astype(dtype)
test_utils.assert_allclose_according_to_type(np_sparsemax, tf_sparsemax_out)

self.assertAllCloseAccordingToType(np_sparsemax, tf_sparsemax_out)
self.assertShapeEqual(np_sparsemax, tf_sparsemax_op)

def test_sparsemax_against_numpy_high_rank(self, dtype=None):
"""check sparsemax kernel against numpy."""
random = np.random.RandomState(1)
@pytest.mark.parametrize("dtype", ["float32", "float64"])
def test_sparsemax_against_numpy_high_rank(dtype):
"""check sparsemax kernel against numpy."""
random = np.random.RandomState(1)

z = random.uniform(low=-3, high=3, size=(test_obs, test_obs, 10))
z = random.uniform(low=-3, high=3, size=(test_obs, test_obs, 10))

tf_sparsemax_op, tf_sparsemax_out = self._tf_sparsemax(z, dtype)
np_sparsemax = np.reshape(
_np_sparsemax(np.reshape(z, [test_obs * test_obs, 10])),
[test_obs, test_obs, 10],
).astype(dtype)
tf_sparsemax_out = sparsemax(z.astype(dtype))
np_sparsemax = np.reshape(
_np_sparsemax(np.reshape(z, [test_obs * test_obs, 10])),
[test_obs, test_obs, 10],
).astype(dtype)

self.assertAllCloseAccordingToType(np_sparsemax, tf_sparsemax_out)
self.assertShapeEqual(np_sparsemax, tf_sparsemax_op)
test_utils.assert_allclose_according_to_type(np_sparsemax, tf_sparsemax_out)

def test_sparsemax_of_nan(self, dtype=None):
"""check sparsemax transfers nan."""
z_nan = np.asarray(
[[0, np.nan, 0], [0, np.nan, np.nan], [np.nan, np.nan, np.nan],]
).astype(dtype)

_, tf_sparsemax_nan = self._tf_sparsemax(z_nan, dtype)
self.assertAllEqual(
@pytest.mark.parametrize("dtype", ["float32", "float64"])
def test_sparsemax_of_nan(dtype):
"""check sparsemax transfers nan."""
z_nan = np.asarray(
[[0, np.nan, 0], [0, np.nan, np.nan], [np.nan, np.nan, np.nan],]
).astype(dtype)

tf_sparsemax_nan = sparsemax(z_nan)
np.testing.assert_equal(
np.array(
[
[np.nan, np.nan, np.nan],
[np.nan, np.nan, np.nan],
[np.nan, np.nan, np.nan],
],
tf_sparsemax_nan,
)
]
),
tf_sparsemax_nan,
)

def test_sparsemax_of_inf(self, dtype=None):
"""check sparsemax is infinity safe."""
z_neg = np.asarray(
[[0, -np.inf, 0], [0, -np.inf, -np.inf], [-np.inf, -np.inf, -np.inf],]
).astype(dtype)
z_pos = np.asarray(
[[0, np.inf, 0], [0, np.inf, np.inf], [np.inf, np.inf, np.inf]]
).astype(dtype)
z_mix = np.asarray(
[[0, np.inf, 0], [0, np.inf, -np.inf], [-np.inf, np.inf, -np.inf]]
).astype(dtype)

_, tf_sparsemax_neg = self._tf_sparsemax(z_neg, dtype)
self.assertAllEqual(
[[0.5, 0, 0.5], [1, 0, 0], [np.nan, np.nan, np.nan]], tf_sparsemax_neg
)

_, tf_sparsemax_pos = self._tf_sparsemax(z_pos, dtype)
self.assertAllEqual(
@pytest.mark.parametrize("dtype", ["float32", "float64"])
def test_sparsemax_of_inf(dtype):
"""check sparsemax is infinity safe."""
z_neg = np.asarray(
[[0, -np.inf, 0], [0, -np.inf, -np.inf], [-np.inf, -np.inf, -np.inf],]
).astype(dtype)
z_pos = np.asarray(
[[0, np.inf, 0], [0, np.inf, np.inf], [np.inf, np.inf, np.inf]]
).astype(dtype)
z_mix = np.asarray(
[[0, np.inf, 0], [0, np.inf, -np.inf], [-np.inf, np.inf, -np.inf]]
).astype(dtype)

tf_sparsemax_neg = sparsemax(z_neg)
np.testing.assert_equal(
np.array([[0.5, 0, 0.5], [1, 0, 0], [np.nan, np.nan, np.nan]]), tf_sparsemax_neg
)

tf_sparsemax_pos = sparsemax(z_pos)
np.testing.assert_equal(
np.array(
[
[np.nan, np.nan, np.nan],
[np.nan, np.nan, np.nan],
[np.nan, np.nan, np.nan],
],
tf_sparsemax_pos,
)
]
),
tf_sparsemax_pos,
)

_, tf_sparsemax_mix = self._tf_sparsemax(z_mix, dtype)
self.assertAllEqual(
tf_sparsemax_mix = sparsemax(z_mix)
np.testing.assert_equal(
np.array(
[
[np.nan, np.nan, np.nan],
[np.nan, np.nan, np.nan],
[np.nan, np.nan, np.nan],
],
tf_sparsemax_mix,
)
]
),
tf_sparsemax_mix,
)


@pytest.mark.parametrize("dtype", ["float32", "float64"])
def test_sparsemax_of_zero(dtype):
"""check sparsemax proposition 1, part 1."""
z = np.zeros((1, 10))

def test_sparsemax_of_zero(self, dtype=None):
"""check sparsemax proposition 1, part 1."""
z = np.zeros((1, 10))
tf_sparsemax_out = sparsemax(z.astype(dtype))
np_sparsemax = np.ones_like(z, dtype=dtype) / z.size

tf_sparsemax_op, tf_sparsemax_out = self._tf_sparsemax(z, dtype)
np_sparsemax = np.ones_like(z, dtype=dtype) / z.size
test_utils.assert_allclose_according_to_type(np_sparsemax, tf_sparsemax_out)

self.assertAllCloseAccordingToType(np_sparsemax, tf_sparsemax_out)
self.assertShapeEqual(np_sparsemax, tf_sparsemax_op)

def test_sparsemax_of_to_inf(self, dtype=None):
"""check sparsemax proposition 1, part 2."""
random = np.random.RandomState(4)
@pytest.mark.parametrize("dtype", ["float32", "float64"])
def test_sparsemax_of_to_inf(dtype):
"""check sparsemax proposition 1, part 2."""
random = np.random.RandomState(4)

z = random.uniform(low=-3, high=3, size=(test_obs, 10))
z = random.uniform(low=-3, high=3, size=(test_obs, 10))

# assume |A(z)| = 1, as z is continues random
z_sort_arg = np.argsort(z, axis=1)[:, ::-1]
z_sort = np.sort(z, axis=-1)[:, ::-1]
gamma_z = z_sort[:, 0] - z_sort[:, 1]
epsilon = (0.99 * gamma_z * 1).reshape(-1, 1)
# assume |A(z)| = 1, as z is continues random
z_sort_arg = np.argsort(z, axis=1)[:, ::-1]
z_sort = np.sort(z, axis=-1)[:, ::-1]
gamma_z = z_sort[:, 0] - z_sort[:, 1]
epsilon = (0.99 * gamma_z * 1).reshape(-1, 1)

# construct the expected 1_A(z) array
p_expected = np.zeros((test_obs, 10), dtype=dtype)
p_expected[np.arange(0, test_obs), z_sort_arg[:, 0]] = 1
# construct the expected 1_A(z) array
p_expected = np.zeros((test_obs, 10), dtype=dtype)
p_expected[np.arange(0, test_obs), z_sort_arg[:, 0]] = 1

tf_sparsemax_op, tf_sparsemax_out = self._tf_sparsemax((1 / epsilon) * z, dtype)
tf_sparsemax_out = sparsemax(((1 / epsilon) * z).astype(dtype))

self.assertAllCloseAccordingToType(p_expected, tf_sparsemax_out)
self.assertShapeEqual(p_expected, tf_sparsemax_op)
test_utils.assert_allclose_according_to_type(p_expected, tf_sparsemax_out)

def test_constant_add(self, dtype=None):
"""check sparsemax proposition 2."""
random = np.random.RandomState(5)

z = random.uniform(low=-3, high=3, size=(test_obs, 10)).astype(dtype)
c = random.uniform(low=-3, high=3, size=(test_obs, 1)).astype(dtype)
@pytest.mark.parametrize("dtype", ["float32", "float64"])
def test_constant_add(dtype):
"""check sparsemax proposition 2."""
random = np.random.RandomState(5)

_, tf_sparsemax_zpc = self._tf_sparsemax(z + c, dtype)
z = random.uniform(low=-3, high=3, size=(test_obs, 10)).astype(dtype)
c = random.uniform(low=-3, high=3, size=(test_obs, 1)).astype(dtype)

_, tf_sparsemax_z = self._tf_sparsemax(z, dtype)
tf_sparsemax_zpc = sparsemax((z + c))

self.assertAllCloseAccordingToType(
tf_sparsemax_zpc, tf_sparsemax_z, half_atol=5e-3
)
tf_sparsemax_z = sparsemax(z)

test_utils.assert_allclose_according_to_type(
tf_sparsemax_zpc, tf_sparsemax_z, half_atol=5e-3
)


@pytest.mark.parametrize("dtype", ["float32", "float64"])
Expand Down