Failed sampling runs should not be cached #452
Labels
Milestone
Comments
|
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. |
riddell-stan
added a commit
to riddell-stan/httpstan
that referenced
this issue
Dec 13, 2020
Do not save any messages from failed sampling runs. For example, if sampling starts but cannot continue because attempts to find appropriate initial values are rejected, there will be some (logger) messages saved to disk. The presence of these messages confuses the cache machinery, which assumes the presence of messages on disk indicates a successful fit. Deleting them avoids the problem. There remains the problem of retrieving the logger messages associated with the failed fit. It's possible that this is something someone might want to do. Future work might address this. Such a user is also free to use CmdStan. Closes stan-dev#452
riddell-stan
added a commit
that referenced
this issue
Dec 14, 2020
Do not save any messages from failed sampling runs. For example, if sampling starts but cannot continue because attempts to find appropriate initial values are rejected, there will be some (logger) messages saved to disk. The presence of these messages confuses the cache machinery, which assumes the presence of messages on disk indicates a successful fit. Deleting them avoids the problem. There remains the problem of retrieving the logger messages associated with the failed fit. It's possible that this is something someone might want to do. Future work might address this. Such a user is also free to use CmdStan. Closes #452
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The test
test_sampling_initialization_failedillustrates the problem. Or at least I think it does. It fails regularly (but not always) on the second run.The text was updated successfully, but these errors were encountered: