0
$\begingroup$

I fit a Student's t-distribution to some data using the No-U-Turn sampler, and called PyMC3's traceplot function to visualize the posteriors.

What do the different solid, dashed, and dotted lines in the histogram mean? I was expecting to see only one curve.

PyMC3 posteriors

$\endgroup$

1 Answer 1

3
$\begingroup$

Each line represents a different chain, or independent MCMC run, and is useful for diagnostics. You should expect, for example, that if MCMC is working well, the sigma parameter will look the same across all four chains, which it doesn't (more tuning, more sampling, or rewriting the model may help).

When you are computing expectations, you should use all four chains.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.