Hello,
\n\nI'm not partipating on this competition, but I've made kernel to predict tennis match outcomes in ATP tournement where I've done many feature engineering and most importantly for this competiton I calculated the elo rating for each palyer before the begining of the match and used that a feature. As a way to double my training data, I inverted the winner and loser (players) specific columns and putting 1-proba if the feature is a probability (like elo probability) when the columns are inverted.
\n\nI think it possible to calculate the elo rating for a Bascket-ball game if we consider a team as being one person and calculate the elo rating for that team. My Kernel can be found in the following link : https://www.kaggle.com/hakeem/eda-data-prep-modeling-predictions-then-betting
\n\nAbout Elo rating system : https://en.wikipedia.org/wiki/Elo_rating_system
\n\nI hope you'll find it usefull :)
", "messages": [ { "id": "759996", "postDate": "02/29/2020 17:46:24", "content": "Hello,
\n\nI'm not partipating on this competition, but I've made kernel to predict tennis match outcomes in ATP tournement where I've done many feature engineering and most importantly for this competiton I calculated the elo rating for each palyer before the begining of the match and used that a feature. As a way to double my training data, I inverted the winner and loser (players) specific columns and putting 1-proba if the feature is a probability (like elo probability) when the columns are inverted.
\n\nI think it possible to calculate the elo rating for a Bascket-ball game if we consider a team as being one person and calculate the elo rating for that team. My Kernel can be found in the following link : https://www.kaggle.com/hakeem/eda-data-prep-modeling-predictions-then-betting
\n\nAbout Elo rating system : https://en.wikipedia.org/wiki/Elo_rating_system
\n\nI hope you'll find it usefull :)
", "rawMarkdown": "Hello,\n\nI'm not partipating on this competition, but I've made kernel to predict tennis match outcomes in ATP tournement where I've done many feature engineering and most importantly for this competiton I calculated the elo rating for each palyer before the begining of the match and used that a feature. As a way to double my training data, I inverted the winner and loser (players) specific columns and putting 1-proba if the feature is a probability (like elo probability) when the columns are inverted.\n\nI think it possible to calculate the elo rating for a Bascket-ball game if we consider a team as being one person and calculate the elo rating for that team. My Kernel can be found in the following link : [https://www.kaggle.com/hakeem/eda-data-prep-modeling-predictions-then-betting](https://www.kaggle.com/hakeem/eda-data-prep-modeling-predictions-then-betting)\n\nAbout Elo rating system : [https://en.wikipedia.org/wiki/Elo_rating_system](https://en.wikipedia.org/wiki/Elo_rating_system)\n\nI hope you'll find it usefull :)", "votes": null }, { "id": "760061", "postDate": "02/29/2020 19:53:51", "content": "I’ve done that in https://www.kaggle.com/christoffer/simple-k-optimized-elo-model in case you want to compare.
", "rawMarkdown": "I’ve done that in https://www.kaggle.com/christoffer/simple-k-optimized-elo-model in case you want to compare.", "votes": null }, { "id": "760469", "postDate": "03/01/2020 10:39:50", "content": "FYI\nhttps://www.kaggle.com/lpkirwin/fivethirtyeight-elo-ratings
", "rawMarkdown": "FYI\nhttps://www.kaggle.com/lpkirwin/fivethirtyeight-elo-ratings", "votes": null }, { "id": "763329", "postDate": "03/04/2020 11:27:01", "content": "My model is taking a composite approach, combining multiple scores (including one based on my own Elo-style ratings) and running an optimization on the weighting factors. However, my results are in line with Christoffer's (close to 0.6), rather than lkpkirwin's (close to 0.53). I have not included home-court advantage in the regular season results - @Christoffer, do you suspect this is the main difference between your results and lkpkirwins?
", "rawMarkdown": "My model is taking a composite approach, combining multiple scores (including one based on my own Elo-style ratings) and running an optimization on the weighting factors. However, my results are in line with Christoffer's (close to 0.6), rather than lkpkirwin's (close to 0.53). I have not included home-court advantage in the regular season results - @Christoffer, do you suspect this is the main difference between your results and lkpkirwins?", "votes": null }, { "id": "763368", "postDate": "03/04/2020 12:12:48", "content": "I think it's probably a combination of home-court advantage, taking actual scores into account, and allowing ratings to carry over between seasons. (At first glance, it looks as the kernel you linked to uses all regular games for rating calculations, leaking future results into earlier seasons, which could be an advantage.)
\n\nBut I haven't had any time to experiment with this lately, so this is all just speculation on my part.
", "rawMarkdown": "I think it's probably a combination of home-court advantage, taking actual scores into account, and allowing ratings to carry over between seasons. (At first glance, it looks as the kernel you linked to uses all regular games for rating calculations, leaking future results into earlier seasons, which could be an advantage.)\n\nBut I haven't had any time to experiment with this lately, so this is all just speculation on my part.", "votes": null } ], "comments": [ { "id": 760061, "author_name": "christoffer", "author_url": "", "post_date": "02/29/2020 19:53:51", "content": "I’ve done that in https://www.kaggle.com/christoffer/simple-k-optimized-elo-model in case you want to compare.
", "votes": null, "replies": [] }, { "id": 760469, "author_name": "hatunina", "author_url": "", "post_date": "03/01/2020 10:39:50", "content": "FYI\nhttps://www.kaggle.com/lpkirwin/fivethirtyeight-elo-ratings
", "votes": null, "replies": [] }, { "id": 763329, "author_name": "kirrer", "author_url": "", "post_date": "03/04/2020 11:27:01", "content": "My model is taking a composite approach, combining multiple scores (including one based on my own Elo-style ratings) and running an optimization on the weighting factors. However, my results are in line with Christoffer's (close to 0.6), rather than lkpkirwin's (close to 0.53). I have not included home-court advantage in the regular season results - @Christoffer, do you suspect this is the main difference between your results and lkpkirwins?
", "votes": null, "replies": [ { "id": 763368, "author_name": "christoffer", "author_url": "", "post_date": "03/04/2020 12:12:48", "content": "I think it's probably a combination of home-court advantage, taking actual scores into account, and allowing ratings to carry over between seasons. (At first glance, it looks as the kernel you linked to uses all regular games for rating calculations, leaking future results into earlier seasons, which could be an advantage.)
\n\nBut I haven't had any time to experiment with this lately, so this is all just speculation on my part.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "759996": "Hello,\n\nI'm not partipating on this competition, but I've made kernel to predict tennis match outcomes in ATP tournement where I've done many feature engineering and most importantly for this competiton I calculated the elo rating for each palyer before the begining of the match and used that a feature. As a way to double my training data, I inverted the winner and loser (players) specific columns and putting 1-proba if the feature is a probability (like elo probability) when the columns are inverted.\n\nI think it possible to calculate the elo rating for a Bascket-ball game if we consider a team as being one person and calculate the elo rating for that team. My Kernel can be found in the following link : [https://www.kaggle.com/hakeem/eda-data-prep-modeling-predictions-then-betting](https://www.kaggle.com/hakeem/eda-data-prep-modeling-predictions-then-betting)\n\nAbout Elo rating system : [https://en.wikipedia.org/wiki/Elo_rating_system](https://en.wikipedia.org/wiki/Elo_rating_system)\n\nI hope you'll find it usefull :)", "760061": "I’ve done that in https://www.kaggle.com/christoffer/simple-k-optimized-elo-model in case you want to compare.", "760469": "FYI\nhttps://www.kaggle.com/lpkirwin/fivethirtyeight-elo-ratings", "763329": "My model is taking a composite approach, combining multiple scores (including one based on my own Elo-style ratings) and running an optimization on the weighting factors. However, my results are in line with Christoffer's (close to 0.6), rather than lkpkirwin's (close to 0.53). I have not included home-court advantage in the regular season results - @Christoffer, do you suspect this is the main difference between your results and lkpkirwins?", "763368": "I think it's probably a combination of home-court advantage, taking actual scores into account, and allowing ratings to carry over between seasons. (At first glance, it looks as the kernel you linked to uses all regular games for rating calculations, leaking future results into earlier seasons, which could be an advantage.)\n\nBut I haven't had any time to experiment with this lately, so this is all just speculation on my part." }, "source": "meta" }