Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

checking for convergence of MCMC? #38

Open
jbloom opened this issue Sep 9, 2024 · 0 comments
Open

checking for convergence of MCMC? #38

jbloom opened this issue Sep 9, 2024 · 0 comments

Comments

@jbloom
Copy link
Contributor

jbloom commented Sep 9, 2024

I was curious if there was some easy way to check for convergence of the MCMC, since the example doesn't show any convergence plots.

As best I can tell, the PosteriorHandler returned by running fit tracks the value of the model parameters (eg, growth advantages) at each timepoint, so in principle those could all be plotted as a function of MCMC step to check for convergence.

However, I could not figure out how to access the loss function (log likelihood) at each step. I would think ideally plotting that (which is just a one dimensional variable) might be the easiest way to check for convergence.

Is there a way to access that after the fitting to make such plots, or is there some other suggested way to check for convergence?

I especially was unsure what represents a "reasonable" number of steps as the example in the evofr documentation uses 1000 warmup followed by 1000 steps (which actually takes a while), but the Bedford lab pipeline for forecasts-ncov only uses 200 warmup and 200 samples.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant