10.11 Keep calm and keep trying
What should we do when diagnostics show problems? We should try to fix them. Sometimes, PyMC will provide suggestions on what to change. Pay attention to those suggestions, and you will save a lot of debugging time. Here, I have listed a few common actions you could take:
Check for typos or other silly mistakes. It is super common even for experts to make ”silly” mistakes. If you misspell the name of a variable, it is highly likely that the model will not even run. But sometimes the mistake is more subtle, and you still get a syntactically valid model that runs, but with the wrong semantics.
Increase the number of samples. This might help for very mild problems, like when you’re close to the target ESS (or MCSE), or when ^R is slightly higher than 1.01 but not too much.
Remove some samples from the beginning of the trace. When checking a trace plot, you may observe that a few samples from the first few steps have overall higher or...
