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

adjust Prodigy recipe to include new configs #68

Closed
6 tasks done
mnaydan opened this issue Aug 15, 2024 · 4 comments
Closed
6 tasks done

adjust Prodigy recipe to include new configs #68

mnaydan opened this issue Aug 15, 2024 · 4 comments
Assignees

Comments

@mnaydan
Copy link
Collaborator

mnaydan commented Aug 15, 2024

We'll have to decide what settings in the recipe versus environment.

For testing:

  • verify that character highlighting works as desired
  • verify that ignore button is no longer shown
  • verify that a page is only shown to two annotators (sessions); if more occur this could be a case of work stealing which can be verified in the logs
@jerielizabeth
Copy link

Put in recipe for now and add comments noting that this is not long term.

Also of note: turning off feed overlap in Ansible main config.

@laurejt laurejt added the 🗜️ awaiting testing Implemented and ready to be tested label Sep 26, 2024
@mnaydan
Copy link
Collaborator Author

mnaydan commented Sep 26, 2024

The character-level highlighting and ignore button issues look good. But as far as I can tell from the test sessions, it's still serving out everything to everyone. Even after annotations are saved/submitted in two sessions, the other test sessions receive those same pages.

@mnaydan mnaydan added ⚠️tested needs attention Has been through acceptance testing and needs additional work and removed 🗜️ awaiting testing Implemented and ready to be tested labels Sep 26, 2024
@mnaydan
Copy link
Collaborator Author

mnaydan commented Sep 30, 2024

@laurejt thanks for loading it with the smaller test set so we can try to test the two-annotators behavior. I'm still running into weird behavior. I opened up 3 different test sessions and each session was annotating different pages (there was no overlap), and I got a "no more tasks available" message after just 1 or 2 pages.

@laurejt laurejt added 🗜️ awaiting testing Implemented and ready to be tested and removed ⚠️tested needs attention Has been through acceptance testing and needs additional work labels Oct 1, 2024
@laurejt
Copy link
Contributor

laurejt commented Oct 2, 2024

For task routing to work properly when we want more than one annotation per task (i.e. page), we need to specify the session names up front. The distribution of annotations will only be approximately even (routing relies on mods of hashes). Work stealing has the potential to make this more uneven, since tasks can be "stolen" from one session and reassigned to another based on the relative age of the assignment. While this can make the work distribution more uneven, it will let us finish faster. For more on work stealing, see the answer I received from the Prodigy support forum.

For the future annotations rounds and efforts, I'm investigating how we might guarantee 2 or more annotations per task without specifying session names. We can probably achieve this by writing a custom task router routine. That said, I've made a post asking Prodigy support on any recommendations for how to proceed.

@laurejt laurejt closed this as completed Oct 2, 2024
@laurejt laurejt removed the 🗜️ awaiting testing Implemented and ready to be tested label Oct 2, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
Archived in project
Development

No branches or pull requests

3 participants