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

Hardcoded tolerance for score comparison #83

Open
Marvin84 opened this issue Nov 19, 2024 · 2 comments
Open

Hardcoded tolerance for score comparison #83

Marvin84 opened this issue Nov 19, 2024 · 2 comments
Assignees
@curufinwe @Marvin84 @michelwi @SimBe195

Comments

@Marvin84
Copy link
Contributor

Marvin84 commented Nov 19, 2024 •
edited
Loading

When calculating NBest at this point we make sure that the forward and backward passes produce the same score within a certain tolerance.
Depending on the model, one could allow for different tolerances. Can we avoid hardcoding this to 100?

I have the case of bestScore_ (my understanding says this is forward score) equal to 0.228684 and bwdScore of the start node equal to 0.228682. I assume this difference to be negligible, however, with tolerance 100 this is gonna crash.
ADD: tolerance 1000 would already work
@curufinwe
Copy link
Contributor

If you want to make it configurable it's fine with me.

@Marvin84
Copy link
Contributor Author

f you want to make it configurable it's fine with me.

Ok, I do a PR for this. Thanks

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@curufinwe curufinwe

@Marvin84 Marvin84

@michelwi michelwi

@SimBe195 SimBe195

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
4 participants
@curufinwe @Marvin84 @michelwi @SimBe195