def test_ScaledArray_array():
for N in range(3, 5):
for P in range(3, 5):
array = np.random.rand(N, P) + 1
std = np.diag(1 / np.std(array, axis=0))
mu = np.mean(array, axis=0)
sarray = ScaledCenterArray(scale=False, center=False)
sarray.fit(da.array(array))
np.testing.assert_array_almost_equal(array, sarray.array)
np.testing.assert_array_almost_equal(array.T, sarray.T.array)
# With Scale but No Center
# B = AD
b_array = array.dot(std)
sarray = ScaledCenterArray(scale=True, center=False)
sarray.fit(da.array(array))
np.testing.assert_array_almost_equal(b_array, sarray.array)
np.testing.assert_array_almost_equal(b_array.T, sarray.T.array)
# With Center but No Scale:
# B = (A - U)
b_array = array - mu
sarray = ScaledCenterArray(scale=False, center=True)
sarray.fit(da.array(array))
np.testing.assert_array_almost_equal(b_array, sarray.array)
np.testing.assert_array_almost_equal(b_array.T, sarray.T.array)
# With Center and Scale:
# (A - U)'D'D(A - U)x
b_array = (array - mu).dot(std)
sarray = ScaledCenterArray(scale=True, center=True)
sarray.fit(da.array(array))
np.testing.assert_array_almost_equal(b_array, sarray.array)
np.testing.assert_array_almost_equal(b_array.T, sarray.T.array)
def test_ScaledArray_sym_mat_mult():
for N in range(2, 5):
for P in range(2, 5):
array = np.random.rand(N, P) + 1
std = np.diag(1/np.std(array, axis=0))
mu = np.mean(array, axis=0)
for factor in [None, 'n', 'p']:
if factor is None:
f = 1
elif factor == 'n':
f = N
else:
f = P
for K in range(1, 5):
for squeeze in [True, False]:
x = np.random.rand(N, K)
if squeeze:
x = np.squeeze(x)
for fit_x in [x, None]:
# With No Scale or Center
# x = A'Ax
result = array.dot(array.T.dot(x))/f
assert result.shape == x.shape
sarray = ScaledCenterArray(scale=False, center=False, factor=factor)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_equal(result, sarray.sym_mat_mult(x))
# With Scale but No Center
# B = AD
b_array = array.dot(std)
result = b_array.dot(b_array.T.dot(x))/f
assert result.shape == x.shape
sarray = ScaledCenterArray(scale=True, center=False, factor=factor)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray.sym_mat_mult(x))
# With Center but No Scale:
# B = (A - U)
b_array = array - mu
result = b_array.dot(b_array.T.dot(x))/f
sarray = ScaledCenterArray(scale=False, center=True, factor=factor)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray.sym_mat_mult(x))
# With Center and Scale:
# (A - U)'D'D(A - U)x
result = (array - mu).dot(std).dot(std).dot((array - mu).T.dot(x))/f
sarray = ScaledCenterArray(scale=True, center=True, factor=factor)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray.sym_mat_mult(x))
def test_array_shapes():
N, K = 10, 7
array = da.array(np.random.rand(N, K))
sarray = ScaledCenterArray()
sarray.fit(array)
assert sarray.shape == (N, K)
assert sarray.T.shape == (K, N)
def test__getitem__base_array_case2():
for N in range(2, 5):
for P in range(2, 5):
array = da.array(np.random.rand(N , P))
sarray = ScaledCenterArray()
sarray.fit(array)
for i in range(2, N):
for j in range(2, P):
np.testing.assert_array_equal(sarray[0:i, 0:j]._array, array[0:i, 0:j])
def test__getitem__mult_case():
for N in range(3, 5):
for P in range(3, 5):
array = np.random.rand(N, P) + 1
for K in range(1, 5):
for sub_N in range(2, N):
sub_array = array[0:sub_N, :]
std = np.diag(1 / np.std(sub_array, axis=0))
mu = np.mean(sub_array, axis=0)
for squeeze in [True, False]:
x = np.random.rand(P, K)
if squeeze:
x = np.squeeze(x)
for fit_x in [x, None]:
# With No Scale or Center
# x = A'Ax
result = sub_array.dot(x)
sarray = ScaledCenterArray(scale=False, center=False)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray[0:sub_N, :].dot(x))
# With Scale but No Center
# B = AD
b_array = sub_array.dot(std)
result = b_array.dot(x)
sarray = ScaledCenterArray(scale=True, center=False)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray[0:sub_N, :].dot(x))
# With Center but No Scale:
# B = (A - U)
b_array = sub_array - mu
result = b_array.dot(x)
sarray = ScaledCenterArray(scale=False, center=True)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray[0:sub_N, :].dot(x))
# With Center and Scale:
# (A - U)'D'D(A - U)x
b_array = (sub_array - mu).dot(std)
result = b_array.dot(x)
sarray = ScaledCenterArray(scale=True, center=True)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray[0:sub_N, :].dot(x))
def test_array_id():
array = da.array(np.random.rand(10, 7))
x = da.array(np.random.rand(10,5))
sarray = ScaledCenterArray(scale=True, center=True)
sarray.fit(da.array(array), x=x)
sarray_T = sarray.T
assert id(sarray._array) == id(sarray_T._array)
assert id(sarray.center_vector) == id(sarray_T.center_vector)
assert id(sarray._array_moment.scale_matrix) == id(sarray_T._array_moment.scale_matrix)
assert id(sarray._array_moment.sym_scale_matrix) == id(sarray_T._array_moment.sym_scale_matrix)
def test_ScaledArray_T_dot():
for N in range(2, 5):
for P in range(2, 5):
array = np.random.rand(N, P) + 1
std = np.diag(1/np.std(array, axis=0))
mu = np.mean(array, axis=0)
for K in range(1, 5):
for squeeze in [True, False]:
x = np.random.rand(N, K)
if squeeze:
x = np.squeeze(x)
for fit_x in [x, None]:
# With No Scale or Center
# x = A'Ax
result = array.T.dot(x)
sarray = ScaledCenterArray(scale=False, center=False)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray.T.dot(x))
# With Scale but No Center
# B = AD
b_array = array.dot(std).T
result = b_array.dot(x)
sarray = ScaledCenterArray(scale=True, center=False)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray.T.dot(x))
# With Center but No Scale:
# B = (A - U)
b_array = (array - mu).T
result = b_array.dot(x)
sarray = ScaledCenterArray(scale=False, center=True)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray.T.dot(x))
# With Center and Scale:
# (A - U)'D'D(A - U)x
b_array = (array - mu).dot(std).T
result = b_array.dot(x)
sarray = ScaledCenterArray(scale=True, center=True)
sarray.fit(da.array(array), x=fit_x)
np.testing.assert_array_almost_equal(result, sarray.T.dot(x))
def test_array_tranpose_tranpose():
array = da.array(np.random.rand(7, 10))
x = da.array(np.random.rand(10, 5))
sarray = ScaledCenterArray(scale=True, center=True)
sarray.fit(da.array(array))
s_array_T_T = sarray.T.T
assert id(sarray._array) == id(s_array_T_T._array)
assert id(sarray) == id(s_array_T_T)
np.testing.assert_array_equal(sarray.dot(x), s_array_T_T.dot(x))
def test__getitem__base_array_case1():
array = da.array(np.random.rand(3, 3))
sarray = ScaledCenterArray()
sarray.fit(array)
np.testing.assert_array_equal(sarray[:, :]._array, sarray._array)
def test_bad_x():
array = da.array(np.random.rand(3, 3))
x = da.array(np.random.rand(3,3,3))
sarray = ScaledCenterArray()
with pytest.raises(ValueError):
sarray.fit(array, x)