{ "id": 193, "title": "3rd place: Chessmetrics - Variant", "content": "

Dear all,


it was a great competition, thanks a lot.

Here some notes regarding my approach.
\r\n


methodology:
\r\n
 
\r\n
- all coding was \r\ndone using perl
\r\n
- I tested some established basics like elo and chessmetrics \r\nusing the training data
\r\n
- I found chessmetrics a very promising approach - regarding \r\nthe parameters performance, opposition quality, activity, weighting, self \r\nconsistant rating over a time period
\r\n


- I varied the rating formula depending on the \r\nparameters above


my best submission materialized in \r\nusing:
\r\n
 
\r\n
- the complete training data from week 1 to 100, and \r\nnot a subset of these data
\r\n
 

- using the iterative rating \r\nformula:\r\n
 
rating = (weighed_performance + weighed_opponent_rating \r\n+ weighed_tie_to_defined_rating_level + extra_points) / (sum(game_weight) + \r\nweight_opponent_rating + weight_tie_to defined_rating_level)\r\n


weight_tie_to defined_rating_level = \r\n2.5


game_weight  = (1 / (1 + ( 100 - month ) / 48 ) )**2
\r\n
 
\r\n
white_advantage = 27.5
\r\n
 
\r\n
extention = 850
\r\n
 
\r\n
 weighed_performance = sum ( (opponent_rating +- \r\nwhite_advantage + extention * \r\n(result-0.5) ) * game_weight \r\n)
\r\n
 
\r\n\r\n
weight_opponent_rating = \r\n12.5
\r\n
 
\r\n
weighed_opponent_ra ting = weight_opponent_rating * sum( (opponent_rating +- \r\nwhite_adavantage) * (game_weight) ) / \r\nsum(game_weight)
\r\n
 
\r\n
defined_rating_level = \r\n2300
\r\n
 
\r\n
weighed_tie_to_defi ned_rating_level = weight_tie_to defined_rating_level * \r\ndefined_rating_level
\r\n
\r\n
\r\n
 
\r\n
extra_points=24.5
\r\n



- using the prediction \r\nformula:
\r\n
 
\r\n
probability(win_player_white) = ((rating_player_white + \r\nwhite_advantage) - (rating_player_black - white_advantage)) / extention  + 0.5 ; limited to \r\n{0..1}


\r\nPerl source of submission 23.


Meanwhile I also updated my profile on Kaggle \r\n...


Cheers


Martin

", "comments": [ { "id": 1255, "author_name": "Martin Reichert", "author_url": "", "post_date": "2010-11-20T11:27:17", "content": "
", "votes": 0, "replies": [] }, { "id": 1256, "author_name": "Martin Reichert", "author_url": "", "post_date": "2010-11-20T11:27:17", "content": "Would somebody please correct the header of the thread?

It should read

3rd place: Chessmetrics - Variant





... and don't find a measure to do it myself.

Thanks a lot
Martin
", "votes": 0, "replies": [] } ], "comment_count": 2, "messages": [ { "id": 1254, "postDate": "2010-11-20T11:27:17Z", "content": "

Dear all,


it was a great competition, thanks a lot.

Here some notes regarding my approach.
\r\n


methodology:
\r\n
 
\r\n
- all coding was \r\ndone using perl
\r\n
- I tested some established basics like elo and chessmetrics \r\nusing the training data
\r\n
- I found chessmetrics a very promising approach - regarding \r\nthe parameters performance, opposition quality, activity, weighting, self \r\nconsistant rating over a time period
\r\n


- I varied the rating formula depending on the \r\nparameters above


my best submission materialized in \r\nusing:
\r\n
 
\r\n
- the complete training data from week 1 to 100, and \r\nnot a subset of these data
\r\n
 

- using the iterative rating \r\nformula:\r\n
 
rating = (weighed_performance + weighed_opponent_rating \r\n+ weighed_tie_to_defined_rating_level + extra_points) / (sum(game_weight) + \r\nweight_opponent_rating + weight_tie_to defined_rating_level)\r\n


weight_tie_to defined_rating_level = \r\n2.5


game_weight  = (1 / (1 + ( 100 - month ) / 48 ) )**2
\r\n
 
\r\n
white_advantage = 27.5
\r\n
 
\r\n
extention = 850
\r\n
 
\r\n
 weighed_performance = sum ( (opponent_rating +- \r\nwhite_advantage + extention * \r\n(result-0.5) ) * game_weight \r\n)
\r\n
 
\r\n\r\n
weight_opponent_rating = \r\n12.5
\r\n
 
\r\n
weighed_opponent_ra ting = weight_opponent_rating * sum( (opponent_rating +- \r\nwhite_adavantage) * (game_weight) ) / \r\nsum(game_weight)
\r\n
 
\r\n
defined_rating_level = \r\n2300
\r\n
 
\r\n
weighed_tie_to_defi ned_rating_level = weight_tie_to defined_rating_level * \r\ndefined_rating_level
\r\n
\r\n
\r\n
 
\r\n
extra_points=24.5
\r\n



- using the prediction \r\nformula:
\r\n
 
\r\n
probability(win_player_white) = ((rating_player_white + \r\nwhite_advantage) - (rating_player_black - white_advantage)) / extention  + 0.5 ; limited to \r\n{0..1}


\r\nPerl source of submission 23.


Meanwhile I also updated my profile on Kaggle \r\n...


Cheers


Martin

" }, { "id": 1255, "postDate": "2010-11-20T11:27:17Z", "content": "
" }, { "id": 1256, "postDate": "2010-11-20T11:27:17Z", "content": "Would somebody please correct the header of the thread?

It should read

3rd place: Chessmetrics - Variant





... and don't find a measure to do it myself.

Thanks a lot
Martin
" } ], "complete": true, "source": "live", "fetched_at": "2026-10-20T02:26:31Z", "competition_id": 2447, "competition_slug": "chess", "team_id": 1179, "solution_url": "/competitions/chess/writeups/pug-3rd-place-chessmetrics-variant" }