Open nzqo opened 6 years ago
My apologies for the delay, I've been on travel for a bit. Thanks for pointing out this issue, in any case. I'd like to investigate a bit further before I say something wrong, but I think there may indeed be a bit of a problem here.
Considering the mus are not ordered, should one not use the mask to specify the remaining kernel locations around which there is still sampling to be done? In the simplest case, assume that only the last particle location in mu yields an invalid particle location when drawing, then in the current situation the algorithm would resample from a gaussian centered around the first location mu[0], not the last.
Edit: tried to add a fix