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

Issue in calculation of log_probs in Sampler #211

Closed
hyeok9855 opened this issue Nov 8, 2024 · 0 comments
Closed

Issue in calculation of log_probs in Sampler #211

hyeok9855 opened this issue Nov 8, 2024 · 0 comments
Assignees
Labels
help wanted Extra attention is needed

Comments

@hyeok9855
Copy link
Collaborator

This issue arises when I implement the LocalSearchSampler, especially the Metropolis-Hastings (MH) rule.

In short, we need both the log_pbs and log_pfs of each trajectory $\tau_{bwd}=(x\to\cdots s' \cdots \to s_0)$ and $\tau_{recon}=(s_0 \to \cdots s' \cdots \to x')$ for MH (see this!).

There are two issues here.

  1. Trajectory reversion may be needed for simpler computation of log_pfs($\tau_{bwd}$) - Check the related comment
  2. It's hard to calculate log_pbs and log_pfs within current structure (TrajectoryBasedGFlowNet has method get_pfs_and_pbs, but Sampler can't access to it) - Check the related comments.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
help wanted Extra attention is needed
Projects
None yet
Development

No branches or pull requests

1 participant