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
Add automatic convergence-checks #140
Comments
first suggestion is now integrated into the run-model function. |
Just circling back on some outstanding issues. I think for (3), it might make sense to include an auto_converge = TRUE, but have an upper limit for the number of re-runs. We could set that by default, and/or allow the user to set it. I'm happy to take on that task, I'll assign myself here. |
We should also incorporate the posterior package, which estimates an updated version of Rhat. The updated Rhat is much less sensitive to asymmetrical posteriors, such as those that are common for the n-values in these models. Vehtari A., Gelman A., Simpson D., Carpenter B., & Bürkner P. C. (2020). Rank-normalization, folding, and localization: An improved Rhat for assessing convergence of MCMC. Bayesian Analysis. |
@AdamCSmithCWS Good idea re posterior package! Are you planning on adding that in on your end and submitting a PR, or would you like me to cover that? Doesn't matter to me! Also, with that updated way of calculating Rhat, does it now make it less "dangerous" to have that 3rd option be implemented, where the model continues to run until fully converged (again,perhaps up until a certain number of total iterations)? |
I think we should have the run_model function include some automatic checks on, at least, cross-chain convergence.
Initial thoughts on options:
The text was updated successfully, but these errors were encountered: