This discussion post summarizes my experience from last year’s competition. I finished with a Silver Medal.
\n\n*Scores during the tournament:*
\n\nEnd of 1st round (after first 32 games):\nMy score: 0.27585\n50th score: 0.30225\n100th score: 0.31411
\n\nEnd of 2nd round (after first 48 games):\nMy score: 0.34370\n50th score: 0.37243\n100th score: 0.38247
\n\nEnd of 3rd round (after first 56 games):\nMy score: 0.32372\n50th score: 0.34295\n100th score: 0.35672
\n\nAt the end of the competition (after all 63 games):\nMy score: 0.35570\n50th score: 0.36147\n100th score: 0.37425
\n\nIt was surprising that the “field” scores decreased (improved) during the Sweet 16 (i.e. you would expect the matches to become more and more unpredictable as the tournament progresses, but this wasn’t the case during the Sweet 16).
\n\n*Reason behind the interesting trend during the Sweet 16 (credit: Ben Tesch)*
\n\nEvery lower seed won during the Sweet 16. Compare this to the second round when 11th seed Missouri State (only 7+ seeded team that reached the Sweet 16) beat 3rd seed Iowa State 69-60, the field scores jumped by .0700.
\n\n*Qualities of good models:*
\n\nIf your model can detect matches that are actually “55-45” where most other competitors are assuming “50-50”, it will slightly improve your rank. But this is important because if you get this wrong, you will marginally but frequently, lose out on the many non-one-sided matches (i.e. | ln(0.5)+ln(0.5) | < | ln(0.45)+ln(0.55) | ). This is the basics your model must get right.
\n\nIf your model can detect matches that have “some chance of upset” where most other competitors are assuming “almost no chance of upset”, it will shoot you up on the leaderboard, even if just a few of the underdogs win. On the other hand, by nature of the log-loss function, your rank won’t drop too much even if no upsets happen. This is where your model gets the bonuses that can separate you from the middle of the pack.
", "messages": [ { "id": "746470", "postDate": "02/15/2020 03:41:11", "content": "This discussion post summarizes my experience from last year’s competition. I finished with a Silver Medal.
\n\n*Scores during the tournament:*
\n\nEnd of 1st round (after first 32 games):\nMy score: 0.27585\n50th score: 0.30225\n100th score: 0.31411
\n\nEnd of 2nd round (after first 48 games):\nMy score: 0.34370\n50th score: 0.37243\n100th score: 0.38247
\n\nEnd of 3rd round (after first 56 games):\nMy score: 0.32372\n50th score: 0.34295\n100th score: 0.35672
\n\nAt the end of the competition (after all 63 games):\nMy score: 0.35570\n50th score: 0.36147\n100th score: 0.37425
\n\nIt was surprising that the “field” scores decreased (improved) during the Sweet 16 (i.e. you would expect the matches to become more and more unpredictable as the tournament progresses, but this wasn’t the case during the Sweet 16).
\n\n*Reason behind the interesting trend during the Sweet 16 (credit: Ben Tesch)*
\n\nEvery lower seed won during the Sweet 16. Compare this to the second round when 11th seed Missouri State (only 7+ seeded team that reached the Sweet 16) beat 3rd seed Iowa State 69-60, the field scores jumped by .0700.
\n\n*Qualities of good models:*
\n\nIf your model can detect matches that are actually “55-45” where most other competitors are assuming “50-50”, it will slightly improve your rank. But this is important because if you get this wrong, you will marginally but frequently, lose out on the many non-one-sided matches (i.e. | ln(0.5)+ln(0.5) | < | ln(0.45)+ln(0.55) | ). This is the basics your model must get right.
\n\nIf your model can detect matches that have “some chance of upset” where most other competitors are assuming “almost no chance of upset”, it will shoot you up on the leaderboard, even if just a few of the underdogs win. On the other hand, by nature of the log-loss function, your rank won’t drop too much even if no upsets happen. This is where your model gets the bonuses that can separate you from the middle of the pack.
", "rawMarkdown": "This discussion post summarizes my experience from last year’s competition. I finished with a Silver Medal.\n \n****Scores during the tournament:****\n\nEnd of 1st round (after first 32 games):\nMy score: 0.27585\n50th score: 0.30225\n100th score: 0.31411\n\nEnd of 2nd round (after first 48 games):\nMy score: 0.34370\n50th score: 0.37243\n100th score: 0.38247\n\nEnd of 3rd round (after first 56 games):\nMy score: 0.32372\n50th score: 0.34295\n100th score: 0.35672\n\nAt the end of the competition (after all 63 games):\nMy score: 0.35570\n50th score: 0.36147\n100th score: 0.37425\n\nIt was surprising that the “field” scores decreased (improved) during the Sweet 16 (i.e. you would expect the matches to become more and more unpredictable as the tournament progresses, but this wasn’t the case during the Sweet 16).\n\n****Reason behind the interesting trend during the Sweet 16 (credit: Ben Tesch)****\n\nEvery lower seed won during the Sweet 16. Compare this to the second round when 11th seed Missouri State (only 7+ seeded team that reached the Sweet 16) beat 3rd seed Iowa State 69-60, the field scores jumped by .0700.\n \n****Qualities of good models:****\n\nIf your model can detect matches that are actually “55-45” where most other competitors are assuming “50-50”, it will slightly improve your rank. But this is important because if you get this wrong, you will marginally but frequently, lose out on the many non-one-sided matches (i.e. | ln(0.5)+ln(0.5) | < | ln(0.45)+ln(0.55) | ). This is the basics your model must get right.\n\nIf your model can detect matches that have “some chance of upset” where most other competitors are assuming “almost no chance of upset”, it will shoot you up on the leaderboard, even if just a few of the underdogs win. On the other hand, by nature of the log-loss function, your rank won’t drop too much even if no upsets happen. This is where your model gets the bonuses that can separate you from the middle of the pack.", "votes": null }, { "id": "749440", "postDate": "02/18/2020 17:39:33", "content": "Thanks for sharing @kaito510 😄 👍
", "rawMarkdown": "Thanks for sharing @kaito510 😄 👍", "votes": null }, { "id": "749709", "postDate": "02/18/2020 21:40:49", "content": "nice info
", "rawMarkdown": "nice info", "votes": null }, { "id": "750791", "postDate": "02/19/2020 17:34:34", "content": "\"It was surprising that the “field” scores decreased (improved) during the Sweet 16 (i.e. you would expect the matches to become more and more unpredictable as the tournament progresses, but this wasn’t the case during the Sweet 16).\"
\n\nYou're right. Last year during the sweet 16, every lower seed won. There wasn't a single upset which is rare. \nIn the first round the biggest upset was two instances of a 10 beating a 7. \nThe second round, an 11 beat a 3, by far the biggest upset in the tournament. I bet a lot submissions took a beating on that game.
\n\nIt seems there are less upsets in the Women's bracket than Men's. I competed last year as a last moment decision and finished in the top 20%. This year, I'm going to give it more attention.
", "rawMarkdown": "\"It was surprising that the “field” scores decreased (improved) during the Sweet 16 (i.e. you would expect the matches to become more and more unpredictable as the tournament progresses, but this wasn’t the case during the Sweet 16).\"\n\nYou're right. Last year during the sweet 16, every lower seed won. There wasn't a single upset which is rare. \nIn the first round the biggest upset was two instances of a 10 beating a 7. \nThe second round, an 11 beat a 3, by far the biggest upset in the tournament. I bet a lot submissions took a beating on that game. \n\nIt seems there are less upsets in the Women's bracket than Men's. I competed last year as a last moment decision and finished in the top 20%. This year, I'm going to give it more attention.", "votes": null }, { "id": "769282", "postDate": "03/11/2020 18:07:27", "content": "Thanks for your insightful comment. Didn't realise what you mentioned about the Sweet 16 when I posted this, I'm editing this post now with a credit to you.
", "rawMarkdown": "Thanks for your insightful comment. Didn't realise what you mentioned about the Sweet 16 when I posted this, I'm editing this post now with a credit to you.", "votes": null } ], "comments": [ { "id": 749440, "author_name": "mashlyn", "author_url": "", "post_date": "02/18/2020 17:39:33", "content": "Thanks for sharing @kaito510 😄 👍
", "votes": null, "replies": [] }, { "id": 749709, "author_name": "", "author_url": "", "post_date": "02/18/2020 21:40:49", "content": "nice info
", "votes": null, "replies": [] }, { "id": 750791, "author_name": "btesch", "author_url": "", "post_date": "02/19/2020 17:34:34", "content": "\"It was surprising that the “field” scores decreased (improved) during the Sweet 16 (i.e. you would expect the matches to become more and more unpredictable as the tournament progresses, but this wasn’t the case during the Sweet 16).\"
\n\nYou're right. Last year during the sweet 16, every lower seed won. There wasn't a single upset which is rare. \nIn the first round the biggest upset was two instances of a 10 beating a 7. \nThe second round, an 11 beat a 3, by far the biggest upset in the tournament. I bet a lot submissions took a beating on that game.
\n\nIt seems there are less upsets in the Women's bracket than Men's. I competed last year as a last moment decision and finished in the top 20%. This year, I'm going to give it more attention.
", "votes": null, "replies": [ { "id": 769282, "author_name": "kaito510", "author_url": "", "post_date": "03/11/2020 18:07:27", "content": "Thanks for your insightful comment. Didn't realise what you mentioned about the Sweet 16 when I posted this, I'm editing this post now with a credit to you.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "746470": "This discussion post summarizes my experience from last year’s competition. I finished with a Silver Medal.\n \n****Scores during the tournament:****\n\nEnd of 1st round (after first 32 games):\nMy score: 0.27585\n50th score: 0.30225\n100th score: 0.31411\n\nEnd of 2nd round (after first 48 games):\nMy score: 0.34370\n50th score: 0.37243\n100th score: 0.38247\n\nEnd of 3rd round (after first 56 games):\nMy score: 0.32372\n50th score: 0.34295\n100th score: 0.35672\n\nAt the end of the competition (after all 63 games):\nMy score: 0.35570\n50th score: 0.36147\n100th score: 0.37425\n\nIt was surprising that the “field” scores decreased (improved) during the Sweet 16 (i.e. you would expect the matches to become more and more unpredictable as the tournament progresses, but this wasn’t the case during the Sweet 16).\n\n****Reason behind the interesting trend during the Sweet 16 (credit: Ben Tesch)****\n\nEvery lower seed won during the Sweet 16. Compare this to the second round when 11th seed Missouri State (only 7+ seeded team that reached the Sweet 16) beat 3rd seed Iowa State 69-60, the field scores jumped by .0700.\n \n****Qualities of good models:****\n\nIf your model can detect matches that are actually “55-45” where most other competitors are assuming “50-50”, it will slightly improve your rank. But this is important because if you get this wrong, you will marginally but frequently, lose out on the many non-one-sided matches (i.e. | ln(0.5)+ln(0.5) | < | ln(0.45)+ln(0.55) | ). This is the basics your model must get right.\n\nIf your model can detect matches that have “some chance of upset” where most other competitors are assuming “almost no chance of upset”, it will shoot you up on the leaderboard, even if just a few of the underdogs win. On the other hand, by nature of the log-loss function, your rank won’t drop too much even if no upsets happen. This is where your model gets the bonuses that can separate you from the middle of the pack.", "749440": "Thanks for sharing @kaito510 😄 👍", "749709": "nice info", "750791": "\"It was surprising that the “field” scores decreased (improved) during the Sweet 16 (i.e. you would expect the matches to become more and more unpredictable as the tournament progresses, but this wasn’t the case during the Sweet 16).\"\n\nYou're right. Last year during the sweet 16, every lower seed won. There wasn't a single upset which is rare. \nIn the first round the biggest upset was two instances of a 10 beating a 7. \nThe second round, an 11 beat a 3, by far the biggest upset in the tournament. I bet a lot submissions took a beating on that game. \n\nIt seems there are less upsets in the Women's bracket than Men's. I competed last year as a last moment decision and finished in the top 20%. This year, I'm going to give it more attention.", "769282": "Thanks for your insightful comment. Didn't realise what you mentioned about the Sweet 16 when I posted this, I'm editing this post now with a credit to you." }, "source": "meta" }