Move over Elo - crowdsourcing a new chess rating system | Hacker News Reader

Move over Elo - crowdsourcing a new chess rating system

kaggle.com

21 pointsdatageek17 comments

As someone who has played competitive chess for many years, I am impressed with just how robust and reliable the elo rating system has proven to be.

A 100 point difference between two players suggests that the stronger player will score 66% ((1#wins + 0.5#draws + 0*#losses)/#totalgames) and a 200 point difference will predict the stronger player will score 75%.

I bet a modification of PageRank would be a fun place to start.

I've been participating in a website that uses directed acyclic graphs to rank NFL teams (http://beatpaths.com), but lately I've been thinking about PageRank after reading a paper on it: http://www.ncsu.edu/crsc/reports/ftp/pdf/crsc-tr06-19.pdf

You could probably make similar adjustments to PageRank to make it suitable for chess.

I tried modifications for PageRank for rating soccer teams, and I will say it provided a questionable rating and not very good predictions compared to other rating methods.

I've been playing on FICS recently. I've always been an average player - good enough to beat my friends but usually lose against anyone who has 'studied' the game. However I decided to try a new strategy: knock them out of their opening book at the earliest opportunity. Spectacular results so far: beat a player rated 300 Elo above me in 9 moves! Probably a fluke though, but still it generally seems to give me a much better chance (I don't have the patience to learn the opening theory).

My favorite strategy is also the all in zerg rush. Works wonders even at higher levels.

I played competition chess for a bit I once nearly got a chessboard thrown at me for that trick ;)

As you say, it works wonders and it throws 'book players' of their game like almost nothing else.

Suddenly they have to play instead of regurgitate.

You were most assuredly playing against below average players. Most of the early part of a chess game is spent trying to establish an 'opening line' and can transform several times depending on what moves both players make. So, you as white could set out trying to play the English and find your self ending up with the QGD.

If you don't learn the theory behind the openings (whether by reading or by observation) you will remain a below average player for a long time

Everybody's submitted solutions to this seemed to have an RMSE of 0.794 to 0.808 which seemed a bit odd to all converge in such a narrow range.

I don't think my solution is that clever, but it's gone straight to the top of the leader-board with a 0.70 RMSE.

what was your methodology or general strategy?

Elo system is really cool, when I was a kid, I thought the name Élő (the inventor's name) is meant literally, as it means "live" in Hungarian, so I thought it is just something that is being constantly updated, which it is in fact.

I think that one of the interesting areas for Elo-type ratings is in online video games. Especially where teams are involved. How do you rate a single player based on what they did in a team game?

League of Legends comes to mind with that, their matchmaking system has seen some growing pains for this exact reason.

World of Warcraft did something similar with its arena teams. Don't ask me to quote details, though, or to tell you how successful it was. Others can probably do that better.

I just read a blog post today about Starcraft 2’s rating system: http://www.sirlin.net/blog/2010/7/24/analyzing-starcraft-2s-.... It seems like Microsoft’s TrueSkill system described there might be very applicable to this chess-ranking problem.

at 50 years old - it's probably due for an upgrade!

Instead of being upgraded the ELO system is actually being applied to other sports because of its solid statistical basis.

It is solid and it is easily computable

Reply on news.ycombinator.com