def test_uneven_width():
ds1 = [None, None]
ds2 = [None, None]
ds1[0] = np.random.randn(10, 2)
ds1[1] = np.random.randn(5, 2)
ds2[0] = np.random.randn(10, 4)
ds2[1] = np.random.randn(5, 3)
fu = FeatureUnion(normalize=True)
with assert_raises(ValueError):
fu.fit_transform((ds1, ds2))
def test_uneven_width():
ds1 = [None, None]
ds2 = [None, None]
ds1[0] = np.random.randn(10, 2)
ds1[1] = np.random.randn(5, 2)
ds2[0] = np.random.randn(10, 4)
ds2[1] = np.random.randn(5, 3)
fu = FeatureUnion(normalize=True)
with assert_raises(ValueError):
fu.fit_transform((ds1, ds2))
def test_uneven_len():
with tempdir():
# This doesn't work with py2.6
with dataset('ds1/', 'w', 'dir-npy') as ds1, \
dataset('ds2/', 'w', 'dir-npy') as ds2:
ds1[0] = np.random.randn(10, 2)
ds1[1] = np.random.randn(5, 2)
ds2[0] = np.random.randn(10, 4)
ds2[1] = np.random.randn(10, 4)
# Uneven length!
fu = FeatureUnion(normalize=False)
with assert_raises(ValueError):
fu.fit_transform((ds1, ds2))
def test_uneven_len():
with tempdir():
# This doesn't work with py2.6
with dataset('ds1/', 'w', 'dir-npy') as ds1, \
dataset('ds2/', 'w', 'dir-npy') as ds2:
ds1[0] = np.random.randn(10, 2)
ds1[1] = np.random.randn(5, 2)
ds2[0] = np.random.randn(10, 4)
ds2[1] = np.random.randn(10, 4)
# Uneven length!
fu = FeatureUnion(normalize=False)
with assert_raises(ValueError):
fu.fit_transform((ds1, ds2))
def test_normalize():
ds1 = [np.random.randn(10, 2), np.random.randn(5, 2)]
ds2 = [100 * np.random.randn(10, 3), 100 * np.random.randn(5, 3)]
fu = FeatureUnion(normalize=True)
mds = fu.fit_transform((ds1, ds2))
assert mds[0].shape == (10, 5)
assert mds[1].shape == (5, 5)
combined = np.concatenate(mds[:])
stds = np.std(combined, axis=1)
print(stds)
assert np.max(stds) - np.min(stds) < 2
def test_union():
ds1 = [None, None]
ds2 = [None, None]
ds1[0] = np.random.randn(10, 2)
ds1[1] = np.random.randn(5, 2)
ds2[0] = np.random.randn(10, 4)
ds2[1] = np.random.randn(5, 4)
rs1 = np.sum(ds1[0], axis=1) + np.sum(ds2[0], axis=1)
rs2 = np.sum(ds1[1], axis=1) + np.sum(ds2[1], axis=1)
fu = FeatureUnion(normalize=False)
mds = fu.fit_transform((ds1, ds2))
assert len(mds) == 2
assert mds[0].shape == (10, 6)
assert mds[1].shape == (5, 6)
np.testing.assert_array_almost_equal(np.sum(mds[0], axis=1), rs1)
np.testing.assert_array_almost_equal(np.sum(mds[1], axis=1), rs2)
def test_normalize():
ds1 = [
np.random.randn(10, 2),
np.random.randn(5, 2)
]
ds2 = [
100 * np.random.randn(10, 3),
100 * np.random.randn(5, 3)
]
fu = FeatureUnion(normalize=True)
mds = fu.fit_transform((ds1, ds2))
assert mds[0].shape == (10, 5)
assert mds[1].shape == (5, 5)
combined = np.concatenate(mds[:])
stds = np.std(combined, axis=1)
print(stds)
assert np.max(stds) - np.min(stds) < 2
def test_union():
ds1 = [None, None]
ds2 = [None, None]
ds1[0] = np.random.randn(10, 2)
ds1[1] = np.random.randn(5, 2)
ds2[0] = np.random.randn(10, 4)
ds2[1] = np.random.randn(5, 4)
rs1 = np.sum(ds1[0], axis=1) + np.sum(ds2[0], axis=1)
rs2 = np.sum(ds1[1], axis=1) + np.sum(ds2[1], axis=1)
fu = FeatureUnion(normalize=False)
mds = fu.fit_transform((ds1, ds2))
assert len(mds) == 2
assert mds[0].shape == (10, 6)
assert mds[1].shape == (5, 6)
np.testing.assert_array_almost_equal(np.sum(mds[0], axis=1), rs1)
np.testing.assert_array_almost_equal(np.sum(mds[1], axis=1), rs2)
def test_dataset():
with tempdir():
# This doesn't work with py2.6
with dataset('ds1.h5', 'w', 'hdf5') as ds1, \
dataset('ds2.h5', 'w', 'hdf5') as ds2:
ds1[0] = np.random.randn(10, 2)
ds1[1] = np.random.randn(5, 2)
ds2[0] = np.random.randn(10, 4)
ds2[1] = np.random.randn(5, 4)
# Compare row sums
rs1 = np.sum(ds1[0], axis=1) + np.sum(ds2[0], axis=1)
rs2 = np.sum(ds1[1], axis=1) + np.sum(ds2[1], axis=1)
fu = FeatureUnion(normalize=False)
mds = fu.fit_transform((ds1, ds2))
assert len(mds) == 2
assert mds[0].shape == (10, 6)
assert mds[1].shape == (5, 6)
np.testing.assert_array_almost_equal(np.sum(mds[0], axis=1), rs1)
np.testing.assert_array_almost_equal(np.sum(mds[1], axis=1), rs2)
def test_dataset():
with tempdir():
# This doesn't work with py2.6
with dataset('ds1.h5', 'w', 'hdf5') as ds1, \
dataset('ds2.h5', 'w', 'hdf5') as ds2:
ds1[0] = np.random.randn(10, 2)
ds1[1] = np.random.randn(5, 2)
ds2[0] = np.random.randn(10, 4)
ds2[1] = np.random.randn(5, 4)
# Compare row sums
rs1 = np.sum(ds1[0], axis=1) + np.sum(ds2[0], axis=1)
rs2 = np.sum(ds1[1], axis=1) + np.sum(ds2[1], axis=1)
fu = FeatureUnion(normalize=False)
mds = fu.fit_transform((ds1, ds2))
assert len(mds) == 2
assert mds[0].shape == (10, 6)
assert mds[1].shape == (5, 6)
np.testing.assert_array_almost_equal(np.sum(mds[0], axis=1), rs1)
np.testing.assert_array_almost_equal(np.sum(mds[1], axis=1), rs2)