Covariance of bins not measured should be infinity (not zero)

[andreufont]

I was looking at how Picca does the sub-sampling, and it looks like it assumes a zero variance for bins not measured.

https://github.com/igmhub/picca/blob/02496df79e088332344e2f9322b681d71117ecbf/py/picca/utils.py#L56

I'm not sure this is used anywhere relevant, but I thought I'd flag it here just in case.

[ngbusca]

I agree, but bins that aren't measured aren't used in the fit. Isn't it equivalent? In other words, isn't the inverse covariance of the measured bins equal to the inverse of the covariance where you delete rows and columns of unmeasured separations?

[andreufont]

Hi Nico, welcome back to Picca!

Yes, what you are saying is correct, and I believe this "issue" is not really an issue in our standard analysis. I just wrote it here as a warning, not as something that needs to be fixed soon.