The competition is about to end, and seeing the randomness in our scores, I think there's a good chance of shakeup. Did you experience random fluctuations too ?
", "messages": [ { "id": "2321480", "postDate": "06/28/2023 15:56:29", "content": "The competition is about to end, and seeing the randomness in our scores, I think there's a good chance of shakeup. Did you experience random fluctuations too ?
", "rawMarkdown": "The competition is about to end, and seeing the randomness in our scores, I think there's a good chance of shakeup. Did you experience random fluctuations too ?", "votes": null }, { "id": "2321488", "postDate": "06/28/2023 16:03:23", "content": "I agree. At some point I feel like I lose my cv lb correlation. However, there is nothing we could do at this point. Hope for the best!!!
", "rawMarkdown": "I agree. At some point I feel like I lose my cv lb correlation. However, there is nothing we could do at this point. Hope for the best!!!", "votes": null }, { "id": "2321673", "postDate": "06/28/2023 18:54:57", "content": "Hi @nikhilmishradev, I experienced 0.001 random fluctuation. 0.001 is a 100 move in the ranking! It can be several hundreds in the dense regions (0.700 in particular)
", "rawMarkdown": "Hi @nikhilmishradev, I experienced 0.001 random fluctuation. 0.001 is a 100 move in the ranking! It can be several hundreds in the dense regions (0.700 in particular)", "votes": null }, { "id": "2321693", "postDate": "06/28/2023 19:13:05", "content": "same here
", "rawMarkdown": "same here", "votes": null }, { "id": "2321697", "postDate": "06/28/2023 19:15:21", "content": "Looking at my submissions, I think even 0.705+ scores are not safe
", "rawMarkdown": "Looking at my submissions, I think even 0.705+ scores are not safe", "votes": null }, { "id": "2321731", "postDate": "06/28/2023 20:18:00", "content": "I think there will be a huge shake-up even in high-score areas. People already observe a low cv lb correlation during the middle of the unextended timeline. And with the extra 1.5 months, I feel the public LB is too overfitted while the host currently uses a fresh private hidden test dataset. I don't even have a solution to choose my final submission.
", "rawMarkdown": "I think there will be a huge shake-up even in high-score areas. People already observe a low cv lb correlation during the middle of the unextended timeline. And with the extra 1.5 months, I feel the public LB is too overfitted while the host currently uses a fresh private hidden test dataset. I don't even have a solution to choose my final submission.", "votes": null }, { "id": "2321743", "postDate": "06/28/2023 20:37:14", "content": "All the best, I can see another gold incoming 💪 @nikhilmishradev @tezdhar
", "rawMarkdown": "All the best, I can see another gold incoming 💪 @nikhilmishradev @tezdhar", "votes": null }, { "id": "2321745", "postDate": "06/28/2023 20:42:59", "content": "guys be brave, just select random subs 💪
", "rawMarkdown": "guys be brave, just select random subs 💪", "votes": null }, { "id": "2321746", "postDate": "06/28/2023 20:45:55", "content": "The worst part is, we can't even tell whether our current submissions perform better than older submission on the private set after the API update.
", "rawMarkdown": "The worst part is, we can't even tell whether our current submissions perform better than older submission on the private set after the API update.", "votes": null }, { "id": "2321750", "postDate": "06/28/2023 20:49:10", "content": "i expect std(private LB - public LB) to be over 0.01. This will probably lead to a significant shake-up.
", "rawMarkdown": "i expect std(private LB - public LB) to be over 0.01. This will probably lead to a significant shake-up.", "votes": null }, { "id": "2321757", "postDate": "06/28/2023 21:01:51", "content": "Sure. In the last few days, I tried to blend some of my models. Guess what, there is no way I use my over 200 earlier submissions because they are all 0.46 scores nowðŸ«
", "rawMarkdown": "Sure. In the last few days, I tried to blend some of my models. Guess what, there is no way I use my over 200 earlier submissions because they are all 0.46 scores now🫠", "votes": null }, { "id": "2321826", "postDate": "06/28/2023 23:48:12", "content": "Even more I think
", "rawMarkdown": "Even more I think", "votes": null }, { "id": "2321845", "postDate": "06/29/2023 00:18:04", "content": "Huge shake-up (and down) from the public LB's top tiers. I always felt that public and private test sets have different distribution but never had the gut to follow through. Should have trusted my instinct :) I believe a lot of teams (including ourselves) failed to choose the highest-private-LB-score submissions as well - ours was a 0.002 gap.
\nAnyway, congratulations on the end of this lengthy competition everyeon!
", "rawMarkdown": "Huge shake-up (and down) from the public LB's top tiers. I always felt that public and private test sets have different distribution but never had the gut to follow through. Should have trusted my instinct :) I believe a lot of teams (including ourselves) failed to choose the highest-private-LB-score submissions as well - ours was a 0.002 gap.\n\nAnyway, congratulations on the end of this lengthy competition everyeon!", "votes": null }, { "id": "2321846", "postDate": "06/29/2023 00:18:40", "content": "Seems we missed top 5 due to selecting some based on lb :( (which bit overfits ;best private was 0.704) , but great learning experience and a long run comes to an end :).. congrats to all teams and winners
", "rawMarkdown": "Seems we missed top 5 due to selecting some based on lb :( (which bit overfits ;best private was 0.704) , but great learning experience and a long run comes to an end :).. congrats to all teams and winners", "votes": null }, { "id": "2321848", "postDate": "06/29/2023 00:19:14", "content": "Same here 😀 but a shake down
", "rawMarkdown": "Same here 😀 but a shake down", "votes": null }, { "id": "2321864", "postDate": "06/29/2023 00:38:29", "content": "We all observed your long holding on the top 1. What a bummer.
", "rawMarkdown": "We all observed your long holding on the top 1. What a bummer.", "votes": null }, { "id": "2321876", "postDate": "06/29/2023 00:58:13", "content": "Yes congrats , happens sometime in kaggle when you missed selecting some based on cv but it’s part of the game. But ya bit frustrating missing some top 2/3 selections we will post out solutions though later
", "rawMarkdown": "Yes congrats , happens sometime in kaggle when you missed selecting some based on cv but it’s part of the game. But ya bit frustrating missing some top 2/3 selections we will post out solutions though later", "votes": null } ], "comments": [ { "id": 2321488, "author_name": "yifucai", "author_url": "", "post_date": "06/28/2023 16:03:23", "content": "I agree. At some point I feel like I lose my cv lb correlation. However, there is nothing we could do at this point. Hope for the best!!!
", "votes": null, "replies": [ { "id": 2321693, "author_name": "nikhilmishradev", "author_url": "", "post_date": "06/28/2023 19:13:05", "content": "same here
", "votes": null, "replies": [] } ] }, { "id": 2321673, "author_name": "gehallak", "author_url": "", "post_date": "06/28/2023 18:54:57", "content": "Hi @nikhilmishradev, I experienced 0.001 random fluctuation. 0.001 is a 100 move in the ranking! It can be several hundreds in the dense regions (0.700 in particular)
", "votes": null, "replies": [ { "id": 2321697, "author_name": "nikhilmishradev", "author_url": "", "post_date": "06/28/2023 19:15:21", "content": "Looking at my submissions, I think even 0.705+ scores are not safe
", "votes": null, "replies": [] } ] }, { "id": 2321731, "author_name": "ataraxian", "author_url": "", "post_date": "06/28/2023 20:18:00", "content": "I think there will be a huge shake-up even in high-score areas. People already observe a low cv lb correlation during the middle of the unextended timeline. And with the extra 1.5 months, I feel the public LB is too overfitted while the host currently uses a fresh private hidden test dataset. I don't even have a solution to choose my final submission.
", "votes": null, "replies": [ { "id": 2321746, "author_name": "woprime", "author_url": "", "post_date": "06/28/2023 20:45:55", "content": "The worst part is, we can't even tell whether our current submissions perform better than older submission on the private set after the API update.
", "votes": null, "replies": [ { "id": 2321757, "author_name": "ataraxian", "author_url": "", "post_date": "06/28/2023 21:01:51", "content": "Sure. In the last few days, I tried to blend some of my models. Guess what, there is no way I use my over 200 earlier submissions because they are all 0.46 scores nowðŸ«
", "votes": null, "replies": [] } ] } ] }, { "id": 2321743, "author_name": "nischaydnk", "author_url": "", "post_date": "06/28/2023 20:37:14", "content": "All the best, I can see another gold incoming 💪 @nikhilmishradev @tezdhar
", "votes": null, "replies": [] }, { "id": 2321745, "author_name": "kvlmll", "author_url": "", "post_date": "06/28/2023 20:42:59", "content": "guys be brave, just select random subs 💪
", "votes": null, "replies": [] }, { "id": 2321750, "author_name": "ymatioun", "author_url": "", "post_date": "06/28/2023 20:49:10", "content": "i expect std(private LB - public LB) to be over 0.01. This will probably lead to a significant shake-up.
", "votes": null, "replies": [ { "id": 2321826, "author_name": "nikhilmishradev", "author_url": "", "post_date": "06/28/2023 23:48:12", "content": "Even more I think
", "votes": null, "replies": [] } ] }, { "id": 2321845, "author_name": "hoangnguyen719", "author_url": "", "post_date": "06/29/2023 00:18:04", "content": "Huge shake-up (and down) from the public LB's top tiers. I always felt that public and private test sets have different distribution but never had the gut to follow through. Should have trusted my instinct :) I believe a lot of teams (including ourselves) failed to choose the highest-private-LB-score submissions as well - ours was a 0.002 gap.
\nAnyway, congratulations on the end of this lengthy competition everyeon!
", "votes": null, "replies": [ { "id": 2321848, "author_name": "gauravbrills", "author_url": "", "post_date": "06/29/2023 00:19:14", "content": "Same here 😀 but a shake down
", "votes": null, "replies": [ { "id": 2321864, "author_name": "hoangnguyen719", "author_url": "", "post_date": "06/29/2023 00:38:29", "content": "We all observed your long holding on the top 1. What a bummer.
", "votes": null, "replies": [ { "id": 2321876, "author_name": "gauravbrills", "author_url": "", "post_date": "06/29/2023 00:58:13", "content": "Yes congrats , happens sometime in kaggle when you missed selecting some based on cv but it’s part of the game. But ya bit frustrating missing some top 2/3 selections we will post out solutions though later
", "votes": null, "replies": [] } ] } ] } ] }, { "id": 2321846, "author_name": "gauravbrills", "author_url": "", "post_date": "06/29/2023 00:18:40", "content": "Seems we missed top 5 due to selecting some based on lb :( (which bit overfits ;best private was 0.704) , but great learning experience and a long run comes to an end :).. congrats to all teams and winners
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2321480": "The competition is about to end, and seeing the randomness in our scores, I think there's a good chance of shakeup. Did you experience random fluctuations too ?", "2321488": "I agree. At some point I feel like I lose my cv lb correlation. However, there is nothing we could do at this point. Hope for the best!!!", "2321673": "Hi @nikhilmishradev, I experienced 0.001 random fluctuation. 0.001 is a 100 move in the ranking! It can be several hundreds in the dense regions (0.700 in particular)", "2321693": "same here", "2321697": "Looking at my submissions, I think even 0.705+ scores are not safe", "2321731": "I think there will be a huge shake-up even in high-score areas. People already observe a low cv lb correlation during the middle of the unextended timeline. And with the extra 1.5 months, I feel the public LB is too overfitted while the host currently uses a fresh private hidden test dataset. I don't even have a solution to choose my final submission.", "2321743": "All the best, I can see another gold incoming 💪 @nikhilmishradev @tezdhar", "2321745": "guys be brave, just select random subs 💪", "2321746": "The worst part is, we can't even tell whether our current submissions perform better than older submission on the private set after the API update.", "2321750": "i expect std(private LB - public LB) to be over 0.01. This will probably lead to a significant shake-up.", "2321757": "Sure. In the last few days, I tried to blend some of my models. Guess what, there is no way I use my over 200 earlier submissions because they are all 0.46 scores now🫠", "2321826": "Even more I think", "2321845": "Huge shake-up (and down) from the public LB's top tiers. I always felt that public and private test sets have different distribution but never had the gut to follow through. Should have trusted my instinct :) I believe a lot of teams (including ourselves) failed to choose the highest-private-LB-score submissions as well - ours was a 0.002 gap.\n\nAnyway, congratulations on the end of this lengthy competition everyeon!", "2321846": "Seems we missed top 5 due to selecting some based on lb :( (which bit overfits ;best private was 0.704) , but great learning experience and a long run comes to an end :).. congrats to all teams and winners", "2321848": "Same here 😀 but a shake down", "2321864": "We all observed your long holding on the top 1. What a bummer.", "2321876": "Yes congrats , happens sometime in kaggle when you missed selecting some based on cv but it’s part of the game. But ya bit frustrating missing some top 2/3 selections we will post out solutions though later" }, "source": "meta" }