def test_rank_scale_versus_scipy(A):
data = A.value
# rank the data all at once
output = rank(data)
# check each column versus scipy equivalent
for i in range(data.shape[1]):
feature = data[:, i]
expected = rankdata(feature)
assert np.allclose(expected, output[:, i])
def spearman_pairwise(A):
"""
Calculates the Spearman rank correlation
coefficient between pairs of columns of
the supplied matrix A (similar to
pandas.DataFrame.corr(method='spearman').
Ties are dealt with using the 'average'
method, as described in rank_data.
"""
assert A.ndim > 1
assert A.shape[1] > 1
A_rank = rank(A)
return pearson_pairwise(A_rank)