def test_masked_periodic():
"""Check that masked periodic works
Two point statistics are part correlation and part
normalization. The correlation sums up the number of possible
2-point states. In masked periodic, we assume that vectors going
across the boundary of the structure come back on the other
side. However, a vector landing in the masked area is discarded
(ie not included in the correlation sum). Below, are the hand
computed correlation and normalization. The correct 2point stats
are the correlation divided by the normalization. First, is the
auto-correlation and second is the cross-correlation.
"""
correct_periodic_mask_auto = np.array([[[2, 1, 2], [1, 4, 1], [2, 1, 2]],
[[1, 0, 0], [0, 2, 0], [0, 0, 1]]])
correct_periodic_mask_cross = np.array([[[1, 3, 1], [2, 0, 2], [1, 1, 1]],
[[0, 1, 2], [2, 0, 2], [1, 2, 0]]])
norm_periodic_mask = np.array([[5, 5, 5], [6, 7, 6], [5, 5, 5]])
mask = get_mask()
array = get_array()
# Auto-Correlation
correct = ((correct_periodic_mask_auto /
norm_periodic_mask).round(3).astype(np.float64))
tested = (correlations.two_point_stats(
array, array, mask=mask,
periodic_boundary=True).compute().round(3).astype(np.float64))
assert (correct == tested).all()
# Cross-Correlation
correct = ((correct_periodic_mask_cross /
norm_periodic_mask).round(3).astype(np.float64))
tested = (correlations.two_point_stats(
array, 1 - array, mask=mask,
periodic_boundary=True).compute().round(3).astype(np.float64))
assert (correct == tested).all()
def test_non_periodic():
"""Test that non-periodic works
Two point statistics are part correlation and part
normalization. The correlation sums up the number of possible
2-point states. In non-periodic, we assume that a vector used to
count up 2 point states can only connect two states in the
structure. A vector going outside of the bounds of the structure
is not counted.
Below, are the hand computed correlation and normalization. The
correct 2point stats are the correlation divided by the
normalization. First, is the auto-correlation and second is the
cross-correlation.
"""
correct_nonperiodic_auto = np.array([[[1, 1, 2], [2, 5, 2], [2, 1, 1]],
[[0, 0, 0], [0, 3, 0], [0, 0, 0]]])
correct_nonperiodic_cross = np.array([[[2, 3, 1], [1, 0, 2], [0, 2, 1]],
[[1, 2, 1], [2, 0, 1], [1, 2, 1]]])
norm_nonperiodic = np.array([[4, 6, 4], [6, 9, 6], [4, 6, 4]])
array = get_array()
# Auto-Correlation
correct = (correct_nonperiodic_auto / norm_nonperiodic).round(3).astype(
np.float64)
tested = (correlations.two_point_stats(
array, array,
periodic_boundary=False).compute().round(3).astype(np.float64))
assert (correct == tested).all()
# Cross-Correlation
correct = (correct_nonperiodic_cross / norm_nonperiodic).round(3).astype(
np.float64)
tested = (correlations.two_point_stats(
array, 1 - array,
periodic_boundary=False).compute().round(3).astype(np.float64))
assert (correct == tested).all()
def run_one(size, size_predict, chunk, chunks_predict, periodic_boundary, cutoff):
"""Generic function to test two points stats shape and chunks"""
x_data = np.random.randint(2, size=size)
chunks = (chunk,) + x_data.shape[1:] # pylint: disable=unsubscriptable-object
x_data = da.from_array(x_data, chunks=chunks)
stats = two_point_stats(
x_data, x_data, periodic_boundary=periodic_boundary, cutoff=cutoff
)
assert stats.compute().shape == size_predict
assert stats.chunks == chunks_predict
def test_periodic():
"""Check that periodic still works
The normalization occurs in the two_point_stats function and the
auto-correlation/cross-correlation occur in the cross_correlation
function. Checking that the normalization is properly calculated.
First is the auto-correlation. Second is the cross-correlation.
"""
array = get_array()
correct = ((correlations.cross_correlation(array, array).compute() /
9).round(3).astype(np.float64))
tested = (correlations.two_point_stats(
array, array).compute().round(3).astype(np.float64))
assert (correct == tested).all()
correct = ((correlations.cross_correlation(array, 1 - array).compute() /
9).round(3).astype(np.float64))
tested = (correlations.two_point_stats(
array, 1 - array).compute().round(3).astype(np.float64))
assert (correct == tested).all()
def test_different_sized_arrays():
"""Check that different sized dask arrays are valid masks.
We want to be able to specify the same mask for each sample. We
also want to be able to specify a different mask for each
sample. This validates that both are possible.
"""
array = da.random.random([1000, 3, 3])
mask_same4all = da.random.randint(0, 2, [3, 3])
mask_diff4all = da.random.randint(0, 2, [1000, 3, 3])
correlations.two_point_stats(array, array, mask=mask_same4all)
# The following check fails. Therefore, the current implementation
# only works for one mask for all or different mask for all, which
# is feature rich enough for me.
# correlations.two_point_stats(array, array, mask=mask_same4some)
correlations.two_point_stats(array, array, mask=mask_diff4all)
def test_bools_ints_valid():
"""Some check that boolean and integers are valid masks
A mask could be true and false specifying where there is a
microstructure. However, it could also be any value in the range
$[0,1]$ which specifies the probability a value is correctly
assigned. The mask right now only implements confidence in a
single phase, although idealy it should represent the confidence
in all phases. However, for the use cases where there are 2
phases, a mask with a probability for one phase also completely
describes the confidence in the other phase. Therefore, this
implementation is complete for 2 phases.
"""
mask_int = da.random.randint(0, 2, [1000, 3, 3])
mask_bool = mask_int.copy().astype(bool)
array = da.random.random([1000, 3, 3])
correlations.two_point_stats(array, array, mask=mask_int)
correlations.two_point_stats(array, array, mask=mask_bool)
def test_non_periodic_masking():
"""Check that non-periodic masking works
In non-periodic masking, vectors that go across the boundary or
land in a mask are not included in the sum.
"""
correct_nonperiodic_mask_auto = np.array([[[1, 0, 1], [1, 4, 1], [1, 0,
1]],
[[0, 0, 0], [0, 2, 0], [0, 0,
0]]])
correct_nonperiodic_mask_cross = np.array([[[1, 3, 1], [1, 0, 1],
[0, 1, 0]],
[[0, 1, 1], [1, 0, 1],
[1, 2, 0]]])
norm_nonperiodic_mask = np.array([[2, 4, 3], [4, 7, 4], [3, 4, 2]])
array = get_array()
mask = get_mask()
# Auto-Correlation
np.allclose(
correlations.two_point_stats(array,
array,
mask=mask,
periodic_boundary=False)[:, 1:-1, 1:-1],
correct_nonperiodic_mask_auto / norm_nonperiodic_mask,
)
# Cross-Correlation
np.allclose(
correlations.two_point_stats(array,
1 - array,
mask=mask,
periodic_boundary=False)[:, 1:-1, 1:-1],
correct_nonperiodic_mask_cross / norm_nonperiodic_mask,
)
def test_non_periodic_masking():
"""Check that non-periodic masking works
In non-periodic masking, vectors that go across the boundary or
land in a mask are not included in the sum.
"""
correct_nonperiodic_mask_auto = np.array([[[1, 0, 1], [1, 4, 1], [1, 0,
1]],
[[0, 0, 0], [0, 2, 0], [0, 0,
0]]])
correct_nonperiodic_mask_cross = np.array([[[1, 3, 1], [1, 0, 1],
[0, 1, 0]],
[[0, 1, 1], [1, 0, 1],
[1, 2, 0]]])
norm_nonperiodic_mask = np.array([[2, 4, 3], [4, 7, 4], [3, 4, 2]])
array = get_array()
mask = get_mask()
# Auto-Correlation
correct = ((correct_nonperiodic_mask_auto /
norm_nonperiodic_mask).round(3).astype(np.float64))
tested = (correlations.two_point_stats(
array, array, mask=mask,
periodic_boundary=False).compute().round(3).astype(np.float64))
assert (correct == tested).all()
# Cross-Correlation
correct = ((correct_nonperiodic_mask_cross /
norm_nonperiodic_mask).round(3).astype(np.float64))
tested = (correlations.two_point_stats(
array, 1 - array, mask=mask,
periodic_boundary=False).compute().round(3).astype(np.float64))
assert (correct == tested).all()
