Issues with _filter_zeroes and Corr about pyerrors HOT 4 CLOSED

I guess the practical consequence of not filtering is that one carries more zeroes around in some cases without affecting the error estimate, right? I would be up for removing the filtering. This should not affect backward compatibility up to the relative precision that was filtered.

from pyerrors.

Going one step further, this would make the attribute is_merged obsolete, right? If that is the case we could consider removing it which could result in a bit of a speed-up in some routines. But this requires more thought.

from pyerrors.

I guess the practical consequence of not filtering is that one carries more zeroes around in some cases without affecting the error estimate, right?

Yes, I think this is correct. I thought it was a nice feature to remove these zeroes but the cases where one exactly cancels the contribution of a subset of configs is probably rare compared to the cases where I currently see problems.

from pyerrors.

Going one step further, this would make the attribute is_merged obsolete, right? If that is the case we could consider removing it which could result in a bit of a speed-up in some routines. But this requires more thought.

I am not sure. First of all, I think that this is a nice information to have. Of course this does not mean that we have to carry it around. If one likes to do the filtering manually, the parameter could still be used to identify the chains that need filtering.

from pyerrors.