Neglected to normalize target values馃槪
\nFrom rank 10+ to rank 800+, lesson paid for with blood
", "messages": [ { "id": "2034634", "postDate": "11/18/2022 09:57:53", "content": "Neglected to normalize target values馃槪
\nFrom rank 10+ to rank 800+, lesson paid for with blood
", "rawMarkdown": "Neglected to normalize target values馃槪\n\nFrom rank 10+ to rank 800+, lesson paid for with blood", "votes": null }, { "id": "2034839", "postDate": "11/18/2022 13:08:57", "content": "Thanks for sharing !
\nAny idea what mechanism is behind that ?
", "rawMarkdown": "Thanks for sharing ! \n\nAny idea what mechanism is behind that ?", "votes": null }, { "id": "2035612", "postDate": "11/19/2022 04:12:33", "content": "Might need this too! 馃槶
\nNext time hope will be the opposite direction for you.
Good luck for the next assignment, in such cases, the learning from a mistake is the most valuable lesson @cuizhengliang! All the best!
", "rawMarkdown": "Good luck for the next assignment, in such cases, the learning from a mistake is the most valuable lesson @cuizhengliang! All the best!", "votes": null }, { "id": "2038260", "postDate": "11/21/2022 09:38:49", "content": "I too dropped nearly 800 ranks.馃槶
\nI standardized the input features for each subset of dates to increase robustness to date variation, but this seems to have created a divergence between public and private.
\nI adopted this approach because the public score increased, but I needed to think a bit more about the CV strategy.
I moved up 700+ ranks. After two submissions (one for test) 10 days before deadline, my rank was around 750, so I give up wondering it would ever work for the remaining days. With hindersight, the public LB was a decoy I should pay more attention on CV which supposed to imitate the priv LB. @cuizhengliang, @moritake04 we share some regrets on either direction.
", "rawMarkdown": "I moved up 700+ ranks. After two submissions (one for test) 10 days before deadline, my rank was around 750, so I give up wondering it would ever work for the remaining days. With hindersight, the public LB was a decoy I should pay more attention on CV which supposed to imitate the priv LB. @cuizhengliang, @moritake04 we share some regrets on either direction.", "votes": null }, { "id": "2041680", "postDate": "11/24/2022 05:58:31", "content": "I'm so sorry to hear that. I also experience a terrible unlucky shake but not huge like yours. Add oil in the next competition and may more good luck with you in the future. 鎰垮ソ杩愪笌浣犲父浼达紒
", "rawMarkdown": "I'm so sorry to hear that. I also experience a terrible unlucky shake but not huge like yours. Add oil in the next competition and may more good luck with you in the future. 鎰垮ソ杩愪笌浣犲父浼达紒", "votes": null }, { "id": "2041884", "postDate": "11/24/2022 09:08:55", "content": "濂界溂鐔熺殑id馃槃
", "rawMarkdown": "濂界溂鐔熺殑id馃槃", "votes": null }, { "id": "2077284", "postDate": "12/27/2022 12:22:53", "content": "Hi, @zeemeen ! Can you describe your solution a bit? E.g. what CV strategy and model did you use?
", "rawMarkdown": "Hi, @zeemeen ! Can you describe your solution a bit? E.g. what CV strategy and model did you use?", "votes": null }, { "id": "2082095", "postDate": "01/01/2023 05:21:37", "content": "After embedding training and test set together, I measure density of test data around each training data and assign training weight inversely porportional to it.
", "rawMarkdown": "After embedding training and test set together, I measure density of test data around each training data and assign training weight inversely porportional to it.", "votes": null } ], "comments": [ { "id": 2034839, "author_name": "alexandervc", "author_url": "", "post_date": "11/18/2022 13:08:57", "content": "Thanks for sharing !
\nAny idea what mechanism is behind that ?
", "votes": null, "replies": [] }, { "id": 2035612, "author_name": "something4kag", "author_url": "", "post_date": "11/19/2022 04:12:33", "content": "Might need this too! 馃槶
\nNext time hope will be the opposite direction for you.
Good luck for the next assignment, in such cases, the learning from a mistake is the most valuable lesson @cuizhengliang! All the best!
", "votes": null, "replies": [] }, { "id": 2038260, "author_name": "moritake04", "author_url": "", "post_date": "11/21/2022 09:38:49", "content": "I too dropped nearly 800 ranks.馃槶
\nI standardized the input features for each subset of dates to increase robustness to date variation, but this seems to have created a divergence between public and private.
\nI adopted this approach because the public score increased, but I needed to think a bit more about the CV strategy.
I moved up 700+ ranks. After two submissions (one for test) 10 days before deadline, my rank was around 750, so I give up wondering it would ever work for the remaining days. With hindersight, the public LB was a decoy I should pay more attention on CV which supposed to imitate the priv LB. @cuizhengliang, @moritake04 we share some regrets on either direction.
", "votes": null, "replies": [ { "id": 2077284, "author_name": "shitovvladimir", "author_url": "", "post_date": "12/27/2022 12:22:53", "content": "Hi, @zeemeen ! Can you describe your solution a bit? E.g. what CV strategy and model did you use?
", "votes": null, "replies": [ { "id": 2082095, "author_name": "zeemeen", "author_url": "", "post_date": "01/01/2023 05:21:37", "content": "After embedding training and test set together, I measure density of test data around each training data and assign training weight inversely porportional to it.
", "votes": null, "replies": [] } ] } ] }, { "id": 2041680, "author_name": "songqizhou", "author_url": "", "post_date": "11/24/2022 05:58:31", "content": "I'm so sorry to hear that. I also experience a terrible unlucky shake but not huge like yours. Add oil in the next competition and may more good luck with you in the future. 鎰垮ソ杩愪笌浣犲父浼达紒
", "votes": null, "replies": [ { "id": 2041884, "author_name": "cuizhengliang", "author_url": "", "post_date": "11/24/2022 09:08:55", "content": "濂界溂鐔熺殑id馃槃
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2034634": "Neglected to normalize target values馃槪\n\nFrom rank 10+ to rank 800+, lesson paid for with blood", "2034839": "Thanks for sharing ! \n\nAny idea what mechanism is behind that ?", "2035612": "Might need this too! 馃槶 \nNext time hope will be the opposite direction for you.", "2036125": "Good luck for the next assignment, in such cases, the learning from a mistake is the most valuable lesson @cuizhengliang! All the best!", "2038260": "I too dropped nearly 800 ranks.馃槶\n\nI standardized the input features for each subset of dates to increase robustness to date variation, but this seems to have created a divergence between public and private.\nI adopted this approach because the public score increased, but I needed to think a bit more about the CV strategy.", "2041521": "I moved up 700+ ranks. After two submissions (one for test) 10 days before deadline, my rank was around 750, so I give up wondering it would ever work for the remaining days. With hindersight, the public LB was a decoy I should pay more attention on CV which supposed to imitate the priv LB. @cuizhengliang, @moritake04 we share some regrets on either direction.", "2041680": "I'm so sorry to hear that. I also experience a terrible unlucky shake but not huge like yours. Add oil in the next competition and may more good luck with you in the future. 鎰垮ソ杩愪笌浣犲父浼达紒", "2041884": "濂界溂鐔熺殑id馃槃", "2077284": "Hi, @zeemeen ! Can you describe your solution a bit? E.g. what CV strategy and model did you use?", "2082095": "After embedding training and test set together, I measure density of test data around each training data and assign training weight inversely porportional to it." }, "source": "meta" }