def test_bigger_matrix(self):
np.random.seed(1234)
nel_row = 20
nel_col = 15
nel_bnd = 10
Z = np.random.randint(100, size=(nel_bnd, nel_row, nel_col))
Z = Z.astype('float64')
ncl_row = 5
ncl_col = 6
ncl_bnd = 3
_, _, row_cl, col_cl, bnd_cl, _ = triclustering(
Z,
ncl_row,
ncl_col,
ncl_bnd,
1.e-5,
100,
1.e-8,
row_clusters_init=np.mod(np.arange(nel_row), ncl_row),
col_clusters_init=np.mod(np.arange(nel_col), ncl_col),
bnd_clusters_init=np.mod(np.arange(nel_bnd), ncl_bnd))
np.testing.assert_array_equal(np.sort(np.unique(row_cl)),
np.arange(ncl_row))
np.testing.assert_array_equal(np.sort(np.unique(col_cl)),
np.arange(ncl_col))
np.testing.assert_array_equal(np.sort(np.unique(bnd_cl)),
np.arange(ncl_bnd))
def test_zero_matrix(self):
# special case for the error - and no nan/inf
Z = np.zeros((8, 9, 10))
ncl_row = 9
ncl_col = 10
ncl_bnd = 8
epsilon = 1.e-6
_, _, _, _, _, error = triclustering(
Z,
ncl_row,
ncl_col,
ncl_bnd,
1.e-5,
100,
epsilon,
row_clusters_init=np.mod(np.arange(9), ncl_row),
col_clusters_init=np.mod(np.arange(10), ncl_col),
bnd_clusters_init=np.mod(np.arange(8), ncl_bnd))
assert np.isclose(error, Z.size * epsilon)
def test_constant_rows_and_cols_matrix(self):
Z = np.stack([np.full((12, 15), i) for i in range(1, 10)])
ncl_row = 12
ncl_col = 15
ncl_bnd = 9
_, _, row_cl, col_cl, bnd_cl, _ = triclustering(
Z,
ncl_row,
ncl_col,
ncl_bnd,
1.e-5,
100,
1.e-8,
row_clusters_init=np.mod(np.arange(12), ncl_row),
col_clusters_init=np.mod(np.arange(15), ncl_col),
bnd_clusters_init=np.mod(np.arange(9), ncl_bnd))
assert np.unique(row_cl).size == 1
assert np.unique(col_cl).size == 1
assert np.unique(bnd_cl).size == ncl_bnd
def test_constant_bnd_matrix(self):
# should give one cluster in bands
Z = np.tile(np.arange(180).reshape((12, 15)), (9, 1, 1))
ncl_row = 12
ncl_col = 15
ncl_bnd = 9
_, _, row_cl, col_cl, bnd_cl, _ = triclustering(
Z,
ncl_row,
ncl_col,
ncl_bnd,
1.e-5,
100,
1.e-8,
row_clusters_init=np.mod(np.arange(12), ncl_row),
col_clusters_init=np.mod(np.arange(15), ncl_col),
bnd_clusters_init=np.mod(np.arange(9), ncl_bnd))
assert np.unique(row_cl).size == ncl_row
assert np.unique(col_cl).size == ncl_col
assert np.unique(bnd_cl).size == 1
def test_as_many_clusters_as_elements(self):
# it should immediately converge (2 iterations)
ncl_row = 8
ncl_col = 7
ncl_bnd = 6
Z = np.random.randint(100, size=(ncl_bnd, ncl_row, ncl_col))
Z = Z.astype('float64')
conv, niterations, _, _, _, _ = triclustering(
Z,
ncl_row,
ncl_col,
ncl_bnd,
1.e-5,
100,
1.e-8,
row_clusters_init=np.mod(np.arange(ncl_row), ncl_row),
col_clusters_init=np.mod(np.arange(ncl_col), ncl_col),
bnd_clusters_init=np.mod(np.arange(ncl_bnd), ncl_bnd))
assert conv
assert niterations == 2
def test_constant_row_matrix(self):
# should give one cluster in rows
Z = np.repeat(np.arange(135).reshape((9, 15)), 12, axis=0)
Z = Z.reshape((9, 12, 15))
ncl_row = 12
ncl_col = 15
ncl_bnd = 9
_, _, row_cl, col_cl, bnd_cl, _ = triclustering(
Z,
ncl_row,
ncl_col,
ncl_bnd,
1.e-5,
100,
1.e-8,
row_clusters_init=np.mod(np.arange(12), ncl_row),
col_clusters_init=np.mod(np.arange(15), ncl_col),
bnd_clusters_init=np.mod(np.arange(9), ncl_bnd))
assert np.unique(row_cl).size == 1
assert np.unique(col_cl).size == ncl_col
assert np.unique(bnd_cl).size == ncl_bnd
def test_small_matrix(self):
np.random.seed(1234)
Z = np.random.permutation(np.arange(60)).reshape((3, 4, 5))
ncl_row = 3
ncl_col = 4
ncl_bnd = 2
conv, niterations, row_cl, col_cl, bnd_cl, error = triclustering(
Z,
ncl_row,
ncl_col,
ncl_bnd,
1.e-5,
100,
1.e-8,
row_clusters_init=[1, 2, 0, 1],
col_clusters_init=[1, 0, 3, 2, 0],
bnd_clusters_init=[1, 0, 0])
assert conv
assert niterations == 2
np.testing.assert_array_equal(row_cl, np.array([1, 2, 0, 1]))
np.testing.assert_array_equal(col_cl, np.array([1, 0, 3, 2, 0]))
np.testing.assert_array_equal(bnd_cl, np.array([1, 0, 0]))
np.testing.assert_almost_equal(error, -4249.966724020571)