Estimate total time when using pymc to sample


Could you post your code in a code-block by putting it as text, between triple backticks (```)? It’s very hard to read screenshots, and impossible to try to re-create problems on our end.

The SMC sampler will evaluate the logp more times that you think, because it does some metropolis steps at each iterator to perturb the particles being used to estimate the posterior (see this tutorial for high-level explaination, or dig into the mutate step of the MH kernel to see what’s going on).

For the multi-cores, see this discussion. The current recommended solution is to split off your likelihood Op into a separate file and import it into your notebook – assuming the problem is that you get the pickle error when cores > 1. If it’s actually a problem with parallel calls to the external library, there’s not really anything that we can do about that on the PyMC side.