Suppress p-value warning for NaNs about climpred HOT 7 CLOSED

maybe we just leave this. probably this should rather be addressed in xskillscore

from climpred.

When running pytest we quite a few warnings. Should get rid of those if possible. Some of them are I guess only based on some \s for latex style docstrings but others come from the metrics.

from climpred.

very often I get this kind of warning. I guess it happens always when we have a nan masked grid cell and then divide by that. also happens for pearson_r and all the compute skills

from climpred.

this Nan warning happens inside xskillscore.pearson_r when applying it to masked values (land grid cells for ocean output.)

from climpred.

There's a suggestion here for how to handle it: https://stackoverflow.com/questions/29688168/mean-nanmean-and-warning-mean-of-empty-slice

This suggests to only wrap places you know you'll get the warning, e.g.

# I expect to see RuntimeWarnings in this block
with warnings.catch_warnings():
    warnings.simplefilter("ignore", category=RuntimeWarning)
    foo = np.nanmean(x, axis=1)

Although I could see this becoming messy to put everywhere. Or perhaps we have a decorator we can put around pearson_r and other stats wrappers.

from climpred.

a nicer and also probabily more performant way would be to only apply the metric when the inputs are not nan. the result is also nan, but to mask the area where the metric will be applied. dont know how thats possible but we did something similar for predictability_horizon

from climpred.

You could replace NaNs with -999 or something, then after the fact re-replace them with NaN.

from climpred.