![\"\"](\"https://media.giphy.com/media/iIGJ0EfRPr0w4hToyB/giphy.gif\")
What an exciting end to the competition!
\nCongrats to all the top placing teams and to everyone who learned something new! Looking forward to reading the solution write-ups.
\nSpecial shout out to @mmotoki @zidmie, @vicensgaitan, @mamasinkgs and @ouranos for becoming competitions GM! (let me know if I missed anyone)
", "messages": [ { "id": "1312255", "postDate": "05/18/2021 00:11:07", "content": "
What an exciting end to the competition!
\nCongrats to all the top placing teams and to everyone who learned something new! Looking forward to reading the solution write-ups.
\nSpecial shout out to @mmotoki @zidmie, @vicensgaitan, @mamasinkgs and @ouranos for becoming competitions GM! (let me know if I missed anyone)
", "rawMarkdown": "\n\nWhat an exciting end to the competition!\n\nCongrats to all the top placing teams and to everyone who learned something new! Looking forward to reading the solution write-ups.\n\nSpecial shout out to @mmotoki @zidmie, @vicensgaitan, @mamasinkgs and @ouranos for becoming competitions GM! (let me know if I missed anyone)", "votes": null }, { "id": "1312256", "postDate": "05/18/2021 00:13:28", "content": "The disparity in score range is actually kind of insane compared to other competitions where people fight for one small 1e-3 metric improvement.
\nLooking forward to what preprocessing and processing approaches were proposed here!
", "rawMarkdown": "The disparity in score range is actually kind of insane compared to other competitions where people fight for one small 1e-3 metric improvement.\n\nLooking forward to what preprocessing and processing approaches were proposed here!", "votes": null }, { "id": "1312259", "postDate": "05/18/2021 00:17:21", "content": "Yep, moreover most of us heavily overfitted.
", "rawMarkdown": "Yep, moreover most of us heavily overfitted.", "votes": null }, { "id": "1312268", "postDate": "05/18/2021 00:28:20", "content": "You missed the hero of this competition @mamasinkgs. He is also going to be a competition GM.
", "rawMarkdown": "You missed the hero of this competition @mamasinkgs. He is also going to be a competition GM.", "votes": null }, { "id": "1312276", "postDate": "05/18/2021 00:38:56", "content": "Thanks for pointing that out. Edited my post and added him. Lots of strong GMs earned from this competition!
", "rawMarkdown": "Thanks for pointing that out. Edited my post and added him. Lots of strong GMs earned from this competition!", "votes": null }, { "id": "1312338", "postDate": "05/18/2021 02:06:43", "content": "Having a higher score on the private than on public doesn’t mean you are necessarily overfitting. The trajectories in the private set are fundamentally harder, which would look more like an additive shift rather than a multiplicative one when comparing public and private scores. Also, it seems like floor prediction was hard for the private set. We made 102 incorrect waypoint floor predictions (0.18 score penalty).
", "rawMarkdown": "Having a higher score on the private than on public doesn’t mean you are necessarily overfitting. The trajectories in the private set are fundamentally harder, which would look more like an additive shift rather than a multiplicative one when comparing public and private scores. Also, it seems like floor prediction was hard for the private set. We made 102 incorrect waypoint floor predictions (0.18 score penalty).", "votes": null }, { "id": "1312771", "postDate": "05/18/2021 08:32:25", "content": "How do you know how much incorrect floor predictions you made? Is there ground truth available somewhere?
", "rawMarkdown": "How do you know how much incorrect floor predictions you made? Is there ground truth available somewhere?", "votes": null }, { "id": "1313516", "postDate": "05/18/2021 15:39:52", "content": "You can make a late submit with different floor predictions. If you do all floors +1, and the score goes up by 15, then you know that your floors are not an underestimate for example.
", "rawMarkdown": "You can make a late submit with different floor predictions. If you do all floors +1, and the score goes up by 15, then you know that your floors are not an underestimate for example.", "votes": null } ], "comments": [ { "id": 1312256, "author_name": "josephjzk", "author_url": "", "post_date": "05/18/2021 00:13:28", "content": "The disparity in score range is actually kind of insane compared to other competitions where people fight for one small 1e-3 metric improvement.
\nLooking forward to what preprocessing and processing approaches were proposed here!
", "votes": null, "replies": [ { "id": 1312259, "author_name": "olaf2000", "author_url": "", "post_date": "05/18/2021 00:17:21", "content": "Yep, moreover most of us heavily overfitted.
", "votes": null, "replies": [] }, { "id": 1312338, "author_name": "tvdwiele", "author_url": "", "post_date": "05/18/2021 02:06:43", "content": "Having a higher score on the private than on public doesn’t mean you are necessarily overfitting. The trajectories in the private set are fundamentally harder, which would look more like an additive shift rather than a multiplicative one when comparing public and private scores. Also, it seems like floor prediction was hard for the private set. We made 102 incorrect waypoint floor predictions (0.18 score penalty).
", "votes": null, "replies": [] }, { "id": 1312771, "author_name": "olaf2000", "author_url": "", "post_date": "05/18/2021 08:32:25", "content": "How do you know how much incorrect floor predictions you made? Is there ground truth available somewhere?
", "votes": null, "replies": [] }, { "id": 1313516, "author_name": "tvdwiele", "author_url": "", "post_date": "05/18/2021 15:39:52", "content": "You can make a late submit with different floor predictions. If you do all floors +1, and the score goes up by 15, then you know that your floors are not an underestimate for example.
", "votes": null, "replies": [] } ] }, { "id": 1312268, "author_name": "mdfahimreshm", "author_url": "", "post_date": "05/18/2021 00:28:20", "content": "You missed the hero of this competition @mamasinkgs. He is also going to be a competition GM.
", "votes": null, "replies": [ { "id": 1312276, "author_name": "robikscube", "author_url": "", "post_date": "05/18/2021 00:38:56", "content": "Thanks for pointing that out. Edited my post and added him. Lots of strong GMs earned from this competition!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1312255": "\n\nWhat an exciting end to the competition!\n\nCongrats to all the top placing teams and to everyone who learned something new! Looking forward to reading the solution write-ups.\n\nSpecial shout out to @mmotoki @zidmie, @vicensgaitan, @mamasinkgs and @ouranos for becoming competitions GM! (let me know if I missed anyone)", "1312256": "The disparity in score range is actually kind of insane compared to other competitions where people fight for one small 1e-3 metric improvement.\n\nLooking forward to what preprocessing and processing approaches were proposed here!", "1312259": "Yep, moreover most of us heavily overfitted.", "1312268": "You missed the hero of this competition @mamasinkgs. He is also going to be a competition GM.", "1312276": "Thanks for pointing that out. Edited my post and added him. Lots of strong GMs earned from this competition!", "1312338": "Having a higher score on the private than on public doesn’t mean you are necessarily overfitting. The trajectories in the private set are fundamentally harder, which would look more like an additive shift rather than a multiplicative one when comparing public and private scores. Also, it seems like floor prediction was hard for the private set. We made 102 incorrect waypoint floor predictions (0.18 score penalty).", "1312771": "How do you know how much incorrect floor predictions you made? Is there ground truth available somewhere?", "1313516": "You can make a late submit with different floor predictions. If you do all floors +1, and the score goes up by 15, then you know that your floors are not an underestimate for example." }, "source": "meta" }