Peter Melchior

I can implement the deflection angles for the NFW halo, so that one gets the new positions by iteratively searching for the observed positions given the pre-lensing positions.

I think there are two ways about that: 1. We stick to post-lensing positions, then a new method, say `getPosition()` would return their pre-lensing positions. That's the easiest way, but...

I ran into this issue today as well, and there's a simple solution for specifying custom bin edges: set `smooth1d` to some small value. The offending line is https://github.com/dfm/corner.py/blob/3ddef76af113a8967114f12c86f1c41292289bc0/src/corner/core.py#L201 because...

The `cutoff` argument is meant to allow for speed-ups in cases of many components. It is unlikely that you will need all components for every sample, so setting e.g. `cutoff=3`...

There shouldn't be errors with `cutoff=None`. Can you post the error and the traceback, please.

Can you post the call of `pygmmis.fit` as well please.

That's not trivial to implement in a general EM algorithm because it would alter the M step. However, one could try something like a proximal projection. That is, after a...

Yes, that can happen. The likelihood for mixture models is multi-modal, so the initialization of the `GMM` plays an important role. You are currently using the default random initialization, and...

It looks like you got some nans in your parameters. Hard to tell from this trace, but I suspect that the k-mean initialization created them. Check the result if the...

The traceback in the notebook shows that the error arises in the `logsumexp` function (line 156 in pygmmis.py). This suggests that there's a numerical over- or under-run. I'm guessing that...