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

Make tau configurable by admin #608

Merged

Conversation

jauggy
Copy link
Member

@jauggy jauggy commented Feb 25, 2025

Tau is configurable by admin in Site Config -> Lobby section

The default value of tau is unchanged (which is 1/3)

Also adding in prevent_sigma_increase to true

@jauggy jauggy marked this pull request as ready for review February 26, 2025 03:29
@L-e-x-o-n
Copy link
Collaborator

Can you please put it under a new Openskill section?

@jauggy
Copy link
Member Author

jauggy commented Feb 28, 2025

I've added it to a Rating section. Calling it more generic just in case we change from openskill to another system.

Also I've added in prevent_sigma_increase to true so that we don't have stuff like this where uncertainty increases:

image

This occurs when tau is very high. The way it works is that before a match is rated, the value of tau is added to uncertainty then the match is rated, so if tau is really high you can gain uncertainty.

    new_rating_groups =
      Enum.map(rating_groups, fn team ->
        Enum.map(team, fn {mu, sigma} ->
          {mu, Math.sqrt(sigma ** 2 + options.tau ** 2)}
        end)
      end)

    output = options.model.rate(new_rating_groups, options)

So your before the match gets rated, your uncertainty becomes Math.sqrt(sigma ** 2 + options.tau ** 2)
Then the match gets rated, and there's a chance uncertainty rises if it doesn't fall enough after rating.

The prevent_sigma_increase is already in the openskill lib.

@L-e-x-o-n L-e-x-o-n merged commit 464d4d3 into beyond-all-reason:master Feb 28, 2025
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants