Error in rollmean(): Parameter 'size' is too large.

[privefl]

E.g. bigutilsr::rollmean(x = 1:3, size = 2).
It expects that (2 * size + 1) > length(x).

What strategy to adopt when it is not the case? (+ recode)

[shadrinams]

Any updates on the issue?

[privefl]

I still don't know what would be the best way to handle this, and whether I should.
Indeed, if the size of the smoothing is too large, you'll get the mean for all values, which is not very useful anyway.

Any better suggestion?

[alexey-larionov]

I came across this error when I tried bed_autoSVD() function with a very small number of variants (I know that normally several thousand of variants should be used; however, occasionally I have to analyse some awkward datasets).

To deal with this error I just explicitly suppress the smoothing log-p-values (as it doesn't make sense in this context) by setting roll.size=0.

On my opinion, it might be practically helpful, if you changed the default behaviour to something similar to what I described above. E.g. change Error to Warning, so the function completes, but warns the user that the smoothing had been suppressed.

[privefl]

Yes, we could default back to 0, or to the largest allowed value, or just implement a version that works even if the size is too large.
I don't really have an opinion on which solution would be the best, this is why I haven't really done anything to solve this yet.