Thread history

From RoboWiki
From Talk:Darkcanuck/RRServer/Ratings
Viewing a history listing
Jump to: navigation, search
descTime User Activity Comment
04:39, 7 October 2011MN (Talk | contribs)New thread created 

It´s been a while, but I finally implemented a Schulze ranking using everything that Schulze paper offers. It has tie-breaking and path reconstruction, together with some statistics and coloring to make easier to track what the method did. It uses pairing APS as ballot.

Tried using % wins in Schulze but it failed miserably as everyone from rank 8 to 817 were tied together, making the system deteriorate into a random ballot ranking. Shadow tied together with Worst with tie-breaking relying on a coin-flip was not what I had in mind. As far as I understood what went wrong, normalizing wins dividing them by the number of battles sacrificed resolvability.

The results can be downloaded here (data using Darkcanuck 2011-08-09 backup): [1] [2] [3] [4]

I also added other ranking systems, like APS/Elo (batch), Premier League and APS that everyone is used to. Also added a Simpson-Kramer ranking, due to Skilgannon request to see who is performing best against each bot. Simpson-Kramer is also good as statistics for Schulze.

I took Chase "for fun" advice seriously and tried to make something more appealing.

MN04:39, 7 October 2011
Personal tools