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.

Sign up for GitHub

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

Jump to bottom

MCMC Initialized chains should include initial point #243

Closed
toastedcrumpets opened this issue Sep 30, 2022 · 1 comment
Closed

MCMC Initialized chains should include initial point #243

toastedcrumpets opened this issue Sep 30, 2022 · 1 comment

Comments

@toastedcrumpets
Copy link
Contributor

During the initialization of the chains for the MCMC optimizer, a Gaussian distribution about an initial point is taken.

ESPEI/espei/optimizers/opt_mcmc.py

Line 98 in 7c79719

chains = rng.normal(tiled_parameters, np.abs(tiled_parameters * std_deviation))

I would suggest including the initial point in that set of initial chains. If everything is set up correctly, this won't matter, but for cases where the standard deviation is too high while the initial guess is quite good, the current behavior will lead to a lot of bad starting points. Modifying the initial set to include the initial guess point should ensure that at least this state (or acceptable permutations of it) will survive the MCMC run. What do you think?
@bocklund
Copy link
Member

Yes, thanks for making this an issue. It's actually a feature I've wanted to implement for awhile, but haven't had the time yet. Feel free to take a stab at it (or anyone else who comes across this issue).

I thought Emcee had a feature to allow giving a 0th iteration or something like that. That could be a way to do it. At worst, modifying the first of the randomly generated chains to match the initial set of parameters should be viable and reproducible.

toastedcrumpets added a commit to toastedcrumpets/ESPEI that referenced this issue Nov 14, 2023
@toastedcrumpets
Add initial/mean point to chains PhasesResearchLab#243
4ca7923
@toastedcrumpets toastedcrumpets mentioned this issue Nov 14, 2023
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@toastedcrumpets @bocklund