def quotient_similarity(m,
partition,
agg='sum',
diag_value=None,
check=False,
n_cpu=DEFAULT_CPUS):
'''Generate quotient similarity matrix based on the given partition of matrix rows
Params:
- m: A symmetric 2D array (can be sparse) containing the similarity matrix
- partition: A list-of-lists partitionning the rows/columns of `m`
- agg: One of ('sum', 'min'. 'max', 'mean', 'getnnz') or a function that takes as parameters the matrix and a range of indices from the partition and aggregates the values across the rows
- check: Logical. Should the dunction check that `partition` is a valid partition of the rows of m? Default is 'True'
Returns a similarity matrix reduced to the dimension induced by the partition
'''
if callable(agg):
f_agg = agg
elif isinstance(agg, str):
if agg in MATRIX_METHOD_STR:
f_agg = lambda m, p: csr_matrix(getattr(m[p, :], agg)(axis=0))
else:
raise ValueError(MATRIX_METHOD_STR_ERR)
else:
raise ValueError(MATRIX_METHOD_STR_ERR)
if check:
if not is_partition(partition, start=0, end=m.shape[0] - 1):
raise ValueError('Please provide a proper partition')
result = merge_row_partition(
merge_row_partition(m, partition, f_agg, n_cpu).T, partition, f_agg,
n_cpu)
if diag_value is not None:
result.setdiag(diag_value)
result.eliminate_zeros()
return result
def test_not_partition2():
p = [[0], [1, 2, 3], [4, 6]]
assert not is_partition(p)
def test_not_partition1():
p = [[0], [1, 2], [4, 5]]
assert not is_partition(p)
def test_not_partition():
p = [[10], [11, 12, 13], [14, 15]]
assert not is_partition(p, start=10, end=16)
def test_is_partition():
p = [[10], [11, 12, 13], [14, 15]]
assert is_partition(p, start=10)
def test_is_partition():
p = [[0], [1, 2, 3], [4, 5]]
assert is_partition(p)