Printable Version of Topic

Posted by: caffeine Nov 8 2017, 05:55 PM

Edit: Recalculated without adding days between tournament due to more accurate figures using this method.

I was curious what it would look like to compute the Elo ratings of players based on historical CTWC win/loss data. I used https://pastebin.com/9W1qejUr for input to the program. If you find any mistakes, please let me know. I thought it would be best to use the https://www.remi-coulom.fr/WHR/ method. It's more computationally expensive, but creates slightly more predictive results. Using the https://github.com/goshrine/whole_history_rating, https://pastebin.com/EYtw48vv I used to compute the ratings. Also, the https://pastebin.com/WwjYuvj6

Generally, Elo ratings need several games before they are accurate. That's the main drawback here. The more tournaments played, and the more times a player makes it to the later rounds, the more accurate his or her ratings should be.

Posted by: wizard944 Nov 9 2017, 03:05 PM

This is cool! I kinda wanna do something similar for guideline tetris. Maybe I'll compile all the TTO results and try and make a rating ladder out of that. Any other tournaments that should be included? Maybe some of the HDO ones.

Posted by: caffeine Oct 23 2018, 03:14 AM

I recomputed the ratings with the 2018 match results added in. https://pastebin.com/raw/ug8sA1zT I used for the program. Please let me know if there are any errors. The format is [winner] vs [loser] [Winner wins = 'W', Winner loses = 'L']. For matches I could not find info on, I used "WLW."

Edit: I ran the algorithm this time with 0 days elapsing between tournaments. It seems to be a more accurate method than adding 365 days between tournaments.
Edit2: Updated with more accurate 2018 round 0 info factored in.

Posted by: XaeL Oct 23 2018, 04:41 AM

Wow, great work caffeine!

I think next year there will be 3-4 new players who will all place top 8, knocking out a lot of the old players.

Posted by: caffeine Oct 6 2019, 09:57 PM

Out of curiosity, I wrote a Monte Carlo simulation for a hypothetical 16-player knock-out tournament. I modeled it using CTWC's format with the Elo ratings derived from the previous 7 CTWC tournaments as well as 2018/2019 Westcoast, Mideast, Eastcoast, Southern, and Desert Qualifiers. Got the data for many of the qualifier events that I couldn't find on video or FB from Chris_F's excellent database on the CTWC discord. https://pastebin.com/2fCJPsbM, and here's the output:

Edit: corrected method for finding probability of a win using WHR and reran the program.
Edit2: reran but with best of 5 matches for top 8.