Hello there,
First of all, I did follow the installation guide from here. I did install on that new conda environment m2w64-toolchain using conda command. I’ve linked my versions of pymc, toolchain and pytensor to this post.
I am running a simple test similar to the Inference Chapter of the beginner’s examples, but with a 2 parameters Weibull distribution, sampling four chains with 1000 draws each. (I’ll put the code if necessary). The goal is only to sample, get posterior distributions and plot trace.
My problem is that when running that code on Spyder, it does not always converge fastly (meaning in less than a few minutes). Running is slow most of the time. I’ve tried to run the same code on a Jupyter notebook multiple times to see if it was linked to my environment set up, or something else and surprisingly there were no slow runs with it (all of them converged under a min).
I do not really understand why Spyder would take time whereas Jupyter would not, so here I am seeking some answers. Spyder being more practical to me, it would be simplier to work with it for later purpose.
I thought about a path access problem, only reason to me why the sampler would take time is that it does not use the gcc compiler even though it’s here. Error is not systematic though, since on few runs, sampler converged fastly. Apart from that, no idea. I tried the same on VSCode, same as for Spyder, very slow sampling.
Any help?