Hi,
thanks for the answers! I am working with pymc 3.11.4 now and I’ll probably wait until an official version of pymc 4 is released. The problem was that I accidentally loaded some packages, among others theano, from outside my conda environment. I now finally have the chains argument and it works fine, so thank you for that.
I have some questions concerning multiprocessing now though. I saw that sample_smc overwrites the number of cores if I only give it one chain. That was not the case in the older pymc version I used before, where one could calculate across multiple cores even with only one SMC chain. Why was it changed? And should I then specify n_cores similar to the number of chains or is it useful to have even more cores?
Also, one more question, although I am not sure you can help me with that. I am working on a linux cluster, and everything works fine when I run an SMC sampler with e.g. 3 chains and 16 cores. When I however submit the same job via condor, I get weird error messages and the trace is not saved. I am not sure if it is a condor or a pymc error, or if something in the cluster setup is not working correctly, so I will also ask our administrator about it, but maybe someone of you has encountered similar problems and is able to help. I’ll paste the error message below.