def test_translate_array():
r"""Test _translate_array with random array."""
array_translated = np.array([[2, 4, 6, 10], [1, 3, 7, 0], [3, 6, 9, 4]])
# Find the means over each dimension
column_means_translated = np.zeros(4)
for i in range(4):
column_means_translated[i] = np.mean(array_translated[:, i])
# Confirm that these means are not all zero
assert (abs(column_means_translated) > 1.e-8).all()
# Compute the origin-centred array
origin_centred_array, _ = _translate_array(array_translated)
# Confirm that the column means of the origin-centred array are all zero
column_means_centred = np.ones(4)
for i in range(4):
column_means_centred[i] = np.mean(origin_centred_array[:, i])
assert (abs(column_means_centred) < 1.e-10).all()
# translating a centered array does not do anything
centred_sphere = 25.25 * np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1],
[-1, 0, 0], [0, -1, 0], [0, 0, -1]])
predicted, _ = _translate_array(centred_sphere)
expected = centred_sphere
assert (abs(predicted - expected) < 1.e-8).all()
# centering a translated unit sphere dose not do anything
shift = np.array([[1, 4, 5], [1, 4, 5], [1, 4, 5], [1, 4, 5], [1, 4, 5],
[1, 4, 5]])
translated_sphere = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1], [-1, 0, 0],
[0, -1, 0], [0, 0, -1]]) + shift
predicted, _ = _translate_array(translated_sphere)
expected = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1], [-1, 0, 0],
[0, -1, 0], [0, 0, -1]])
assert (abs(predicted - expected) < 1.e-8).all()
# If an arbitrary array is centroid translated, the analysis applied to the original array
# and the translated array should give identical results
# Define an arbitrary array
array_a = np.array([[1, 5, 7], [8, 4, 6]])
# Define an arbitrary translation
translate = np.array([[5, 8, 9], [5, 8, 9]])
# Define the translated original array
array_translated = array_a + translate
# Begin translation analysis
centroid_a_to_b, _ = _translate_array(array_a, array_translated)
assert (abs(centroid_a_to_b - array_translated) < 1.e-10).all()
def test_scale_array():
r"""Test _scale_array with random array."""
# Rescale arbitrary array
array_a = np.array([[6, 2, 1], [5, 2, 9], [8, 6, 4]])
# Confirm Frobenius normaliation has transformed the array to lie on the unit sphere in
# the R^(mxn) vector space. We must centre the array about the origin before proceeding
array_a, _ = _translate_array(array_a)
# Confirm proper centering
column_means_centred = np.zeros(3)
for i in range(3):
column_means_centred[i] = np.mean(array_a[:, i])
assert (abs(column_means_centred) < 1.e-10).all()
# Proceed with Frobenius normalization
scaled_array, _ = _scale_array(array_a)
# Confirm array has unit norm
assert abs(np.sqrt((scaled_array**2.).sum()) - 1.) < 1.e-10
# This test verifies that when _scale_array is applied to two scaled unit spheres,
# the Frobenius norm of each new sphere is unity.
# Rescale spheres to unitary scale
# Define arbitrarily scaled unit spheres
unit_sphere = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1], [-1, 0, 0],
[0, -1, 0], [0, 0, -1]])
sphere_1 = 230.15 * unit_sphere
sphere_2 = .06 * unit_sphere
# Proceed with scaling procedure
scaled1, _ = _scale_array(sphere_1)
scaled2, _ = _scale_array(sphere_2)
# Confirm each scaled array has unit Frobenius norm
assert abs(np.sqrt((scaled1**2.).sum()) - 1.) < 1.e-10
assert abs(np.sqrt((scaled2**2.).sum()) - 1.) < 1.e-10
# If an arbitrary array is scaled, the scaling analysis should be able to recreate the scaled
# array from the original
# applied to the original array and the scaled array should give identical results.
# Define an arbitrary array
array_a = np.array([[1, 5, 7], [8, 4, 6]])
# Define an arbitrary scaling factor
scale = 6.3
# Define the scaled original array
array_scaled = scale * array_a
# Begin scaling analysis
scaled_a, _ = _scale_array(array_a, array_scaled)
assert (abs(scaled_a - array_scaled) < 1.e-10).all()
# Define an arbitrary array
array_a = np.array([[6., 12., 16., 7.], [4., 16., 17., 33.],
[5., 17., 12., 16.]])
# Define the scaled original array
array_scale = 123.45 * array_a
# Verify the validity of the translate_scale analysis
# Proceed with analysis, matching array_trans_scale to array
predicted, _ = _scale_array(array_scale, array_a)
# array_trans_scale should be identical to array after the above analysis
expected = array_a
assert (abs(predicted - expected) < 1.e-10).all()
def test_translate_weight():
r"""Test _translate_array with weighted array."""
rng = np.random.RandomState(789)
arr = rng.randint(0, 10, (4, 3))
weight = np.arange(1, 5)
# center the data points to mass of center, the way in the notes
arr_weighted = np.dot(np.diag(weight), arr)
col_sum = np.dot(np.ones((arr_weighted.shape[0], arr_weighted.shape[0])), arr_weighted)
# center the data points to mass of center
arr_centered = arr - col_sum / weight.sum()
array_a_centered, _ = _translate_array(array_a=arr,
array_b=None,
weight=weight)
assert_almost_equal(arr_centered, array_a_centered)
