110 lines
3.3 KiB
Python
110 lines
3.3 KiB
Python
|
import numpy as np
|
||
|
|
from numpy.testing import assert_allclose
|
||
|
|
|
||
|
|
import pytest
|
||
|
|
|
||
|
|
import scipy.special as sc
|
||
|
|
|
||
|
|
|
||
|
|
@pytest.mark.parametrize('x, expected', [
|
||
|
|
(np.array([1000, 1]), np.array([0, -999])),
|
||
|
|
|
||
|
|
# Expected value computed using mpmath (with mpmath.mp.dps = 200) and then
|
||
|
|
# converted to float.
|
||
|
|
(np.arange(4), np.array([-3.4401896985611953,
|
||
|
|
-2.4401896985611953,
|
||
|
|
-1.4401896985611953,
|
||
|
|
-0.44018969856119533]))
|
||
|
|
])
|
||
|
|
def test_log_softmax(x, expected):
|
||
|
|
assert_allclose(sc.log_softmax(x), expected, rtol=1e-13)
|
||
|
|
|
||
|
|
|
||
|
|
@pytest.fixture
|
||
|
|
def log_softmax_x():
|
||
|
|
x = np.arange(4)
|
||
|
|
return x
|
||
|
|
|
||
|
|
|
||
|
|
@pytest.fixture
|
||
|
|
def log_softmax_expected():
|
||
|
|
# Expected value computed using mpmath (with mpmath.mp.dps = 200) and then
|
||
|
|
# converted to float.
|
||
|
|
expected = np.array([-3.4401896985611953,
|
||
|
|
-2.4401896985611953,
|
||
|
|
-1.4401896985611953,
|
||
|
|
-0.44018969856119533])
|
||
|
|
return expected
|
||
|
|
|
||
|
|
|
||
|
|
def test_log_softmax_translation(log_softmax_x, log_softmax_expected):
|
||
|
|
# Translation property. If all the values are changed by the same amount,
|
||
|
|
# the softmax result does not change.
|
||
|
|
x = log_softmax_x + 100
|
||
|
|
expected = log_softmax_expected
|
||
|
|
assert_allclose(sc.log_softmax(x), expected, rtol=1e-13)
|
||
|
|
|
||
|
|
|
||
|
|
def test_log_softmax_noneaxis(log_softmax_x, log_softmax_expected):
|
||
|
|
# When axis=None, softmax operates on the entire array, and preserves
|
||
|
|
# the shape.
|
||
|
|
x = log_softmax_x.reshape(2, 2)
|
||
|
|
expected = log_softmax_expected.reshape(2, 2)
|
||
|
|
assert_allclose(sc.log_softmax(x), expected, rtol=1e-13)
|
||
|
|
|
||
|
|
|
||
|
|
@pytest.mark.parametrize('axis_2d, expected_2d', [
|
||
|
|
(0, np.log(0.5) * np.ones((2, 2))),
|
||
|
|
(1, np.array([[0, -999], [0, -999]]))
|
||
|
|
])
|
||
|
|
def test_axes(axis_2d, expected_2d):
|
||
|
|
assert_allclose(
|
||
|
|
sc.log_softmax([[1000, 1], [1000, 1]], axis=axis_2d),
|
||
|
|
expected_2d,
|
||
|
|
rtol=1e-13,
|
||
|
|
)
|
||
|
|
|
||
|
|
|
||
|
|
@pytest.fixture
|
||
|
|
def log_softmax_2d_x():
|
||
|
|
x = np.arange(8).reshape(2, 4)
|
||
|
|
return x
|
||
|
|
|
||
|
|
|
||
|
|
@pytest.fixture
|
||
|
|
def log_softmax_2d_expected():
|
||
|
|
# Expected value computed using mpmath (with mpmath.mp.dps = 200) and then
|
||
|
|
# converted to float.
|
||
|
|
expected = np.array([[-3.4401896985611953,
|
||
|
|
-2.4401896985611953,
|
||
|
|
-1.4401896985611953,
|
||
|
|
-0.44018969856119533],
|
||
|
|
[-3.4401896985611953,
|
||
|
|
-2.4401896985611953,
|
||
|
|
-1.4401896985611953,
|
||
|
|
-0.44018969856119533]])
|
||
|
|
return expected
|
||
|
|
|
||
|
|
|
||
|
|
def test_log_softmax_2d_axis1(log_softmax_2d_x, log_softmax_2d_expected):
|
||
|
|
x = log_softmax_2d_x
|
||
|
|
expected = log_softmax_2d_expected
|
||
|
|
assert_allclose(sc.log_softmax(x, axis=1), expected, rtol=1e-13)
|
||
|
|
|
||
|
|
|
||
|
|
def test_log_softmax_2d_axis0(log_softmax_2d_x, log_softmax_2d_expected):
|
||
|
|
x = log_softmax_2d_x.T
|
||
|
|
expected = log_softmax_2d_expected.T
|
||
|
|
assert_allclose(sc.log_softmax(x, axis=0), expected, rtol=1e-13)
|
||
|
|
|
||
|
|
|
||
|
|
def test_log_softmax_3d(log_softmax_2d_x, log_softmax_2d_expected):
|
||
|
|
# 3-d input, with a tuple for the axis.
|
||
|
|
x_3d = log_softmax_2d_x.reshape(2, 2, 2)
|
||
|
|
expected_3d = log_softmax_2d_expected.reshape(2, 2, 2)
|
||
|
|
assert_allclose(sc.log_softmax(x_3d, axis=(1, 2)), expected_3d, rtol=1e-13)
|
||
|
|
|
||
|
|
|
||
|
|
def test_log_softmax_scalar():
|
||
|
|
assert_allclose(sc.log_softmax(1.0), 0.0, rtol=1e-13)
|