Lukas Hergt

> @williamjameshandley @lukashergt have you noticed this happening with CLASS too? Yes, I initially ran into this with Cobaya+PolyChord+CLASS+Planck runs.

> `clik` done in master. Waiting for updates on the `camb.mathutils.chi_squared` cases. I've had issues with CLASS in the past, any issues with CAMB I managed to fix. So I...

Yes, that is how I have interpreted the `process rank 15` output by the error... That is the parameter set that I've tested directly with Class and with an additional...

## output from cobaya showing `nan` I sprinkled some additional `self.log.debug` statements (in `_planck_clik_prototype.py`'s `log_likelihood` function just before the call to clik) and got some more helpful output: [test_EE_omegak_3d_debug_with_nans.log](https://github.com/CobayaSampler/cobaya/files/4751619/test_EE_omegak_3d_debug_with_nans.log) Search...

Wow! I ran the mini-script again and **this time I got nans!** For the same script I got different output. Here is the output: [lukas_bug2_withnan.log](https://github.com/CobayaSampler/cobaya/files/4752022/lukas_bug2_withnan.log). Running vimdiff on this output...

I've found that I can prevent (or at least heavily reduce) this non-deterministic behaviour (sometimes producing nans and sometimes not) from happening by suppressing all sort of multi-threading, i.e. compiling...

Hi @NiallJeffrey, anesthetic version 1.3.6 is fairly old by now, so my initial guess would be that it might no longer be compatible with newer versions of pandas and/or matplotlib....

Is this resolved by #378?

jupyterlab-vim has moved to https://github.com/jupyterlab-contrib/jupyterlab-vim. No point in adding to issues here...

To better understand the issue, what prompts this? The code not being sufficiently homogeneous or people being annoyed by flake8 complaints after modifications? I have mostly internalised flake8, so I...