spearman returns NaN.

[GregorySchwartz]

spearman should return Nothing rather than NaN:

spearman $ V.fromList [(1, 3), (5, 3)], or any case where one of the two variables is a constant.

[Shimuuar]

Good catch! Would you mind to provide fix for it?

[GregorySchwartz]

This would change correlation, pearson, and pearsonMatByRow to return Maybe as well. pearsonMatByRow may have issues as well if the Matrix type should have Nothing as elements rather than throwing out the entire matrix, because there is no instance for Unbox (Maybe a).

[Shimuuar]

Ohhh... It's going to be a big change. Feel free to share your opinion on the matter

[Shimuuar]

I starting to think that we should do such conversion and return Nothing when no result could be computed. We could do it incrementally and start by convertin correlation & pearson. And decide what to do with pearsonMatByRow later

[GregorySchwartz]

That sounds fine, but maybe return Nothing for pearsonMatByRow? The only issue is when finding bugs, it would be a tad more annoying to see which comparison is returning Nothing.

[ocramz]

I've started addressing this here: https://github.com/DataHaskell/statistics/tree/exceptions-not-error