rdms.rank_transform treats nans as data

[caiw]

When an RDMs.dissimilarities contains nan values, calling rdms.rank_transform on it treats the nans as data, and (by default when using rank_transform(method='average')) assigns them sequential integer ranks.

A better option for me (and perhaps the least surprising) would be if nan values were ignored (i.e. kept as nan) when running rank_transform.

I often pass an RDM through rank_transform when visualising it, like show_rdms(rank_transform(rdms)), and this causes weird colours when an RDM contains a lot of nans (especially when entire rows/columns are missing).

[caiw]

The ability to change how nans are handled by scipy.stats.rankdata was changed in scipy v1.1.0, with the new nan_policy argument, which to achieve the desired solution above should be set to omit.

[JasperVanDenBosch]

Agreed, let's fix this

[caiw]

I've submitted a PR #403, it's a 1-line fix!