Closed
Description
Hi,
in dist_metrics.pyx the correlation metric is commented, and the comment states that it is not a 'true' metric. While this is true as it doesn't satisfy the coincidence axiom, it is still helpful to have such a metric implemented in cython to speed up computations.
Also, note that the actual formula implemented in the commented code is incorrect. The returned value at the moment is
return (1. - x1Tx2) / sqrt(x1nrm * x2nrm)but it should be instead
return 1. - (x1Tx2 / sqrt(x1nrm * x2nrm))Thanks,
Matteo
Metadata
Metadata
Assignees
Labels
No labels