Hi everyone!
\nI wanted to ask a question: Why are there so many people with a score of 28.2?
:D .. Fingers crossed for Bronze
", "rawMarkdown": ":D .. Fingers crossed for Bronze ", "replies": [ { "id": 3221255, "postDate": "2025-06-10T16:37:13.640Z", "content": "Winning bronze in Kaggle: the art of being the first to copy and submit notebooks 🤣
", "rawMarkdown": "Winning bronze in Kaggle: the art of being the first to copy and submit notebooks 🤣", "votes": 6 }, { "id": 3221505, "postDate": "2025-06-11T05:29:40.017Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 3221218, "postDate": "2025-06-10T15:27:13.803Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3221515, "postDate": "2025-06-11T05:41:20.970Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 3221177, "postDate": "2025-06-10T14:08:50.137Z", "content": "Hi everyone!
\nI wanted to ask a question: Why are there so many people with a score of 28.2?
I always wonder, what is the most dramatic public/private ranking jumps observed? For example, if one ranks 1573 on public LB and 102 on private LB, the jump would be +1471.
\nCan some Kagglers share what're the most extreme ranking jumps (both positive and negative) they ever observed?
\nThis is called shakeup, right? Sorry I'm relatively new to Kaggling.
", "rawMarkdown": "I always wonder, what is the most dramatic public/private ranking jumps observed? For example, if one ranks 1573 on public LB and 102 on private LB, the jump would be +1471.\n\nCan some Kagglers share what're the most extreme ranking jumps (both positive and negative) they ever observed?\n\nThis is called shakeup, right? Sorry I'm relatively new to Kaggling.", "replies": [ { "id": 3221766, "postDate": "2025-06-11T12:14:24.387Z", "content": "There are competitions that are pure lottery, with jumps in the thousands. More than a few. I think I saw 4K+ for first place sometime?
\nBut this competition will have very small shake, almost certainly.
https://www.kaggle.com/competitions/blood-vessel-segmentation/leaderboard
\nhttps://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/475657
This is absolutely the most unbelievable ghost story. This player had terrible scores on the public leaderboard and seemingly threw in the towel, but somehow their solution ended up securing second place!
", "rawMarkdown": "https://www.kaggle.com/competitions/blood-vessel-segmentation/leaderboard\nhttps://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/475657\n\nThis is absolutely the most unbelievable ghost story. This player had terrible scores on the public leaderboard and seemingly threw in the towel, but somehow their solution ended up securing second place!", "votes": 3, "replies": [ { "id": 3221986, "postDate": "2025-06-11T17:08:42.250Z", "content": "Go check out AIMO-2 and AIMO-1
\nPeople from top research centers across the world (like ETH Zurich) who ruled the public leaderboard for months had reversed polarity on the private set. Happened with many of the top 10-20. And out of nowhere some people with barely 20-30 submissions popped off to the top places.
In AIMO the very top were stable. If you were really good you win, it was just really really hard to be better than the general noise range.
\nYou want to see real lottery? Check this or this.
\nOh and BELKA was absolutely ridiculous, with 3 out of top 5 novices with 10 or less submissions in the competition.
\nBut we usually can spot lottery competitions long before the shake up, and the opposite is also true. This one, for example, will NOT be lottery.
Even though this competition will not be lottery, surely the countless public 28.2 contestants can't all rank better than all the public 28.3+ contestants in private ranking, right?
\nI wonder how many public 28.3+ contestants would actually beat (at least half of) the public 28.2 contestants in private ranking.
\nAny educated clue?
", "rawMarkdown": "Even though this competition will not be lottery, surely the countless public 28.2 contestants can't all rank better than all the public 28.3+ contestants in private ranking, right?\n\nI wonder how many public 28.3+ contestants would actually beat (at least half of) the public 28.2 contestants in private ranking.\n\nAny educated clue?" }, { "id": 3222075, "postDate": "2025-06-11T19:16:12.723Z", "content": "Sure. A lot of competitions are very stable at the top and shaky down the line.
\nThis is already not due to 'lottery' of conpetition, but because it's a hard one so most people can't get good and well validated solutions that will separate them from the noise of the public (in this competition-28.2 is noise range since it's just copy public notebook and maybe change a few parameters).
\nAs long as top solutions are not lottery, its good.
I think it's because we copy the notebook published in Code, run it, and create submission.csv
", "rawMarkdown": "I think it's because we copy the notebook published in Code, run it, and create submission.csv" } ], "comments": [ { "id": 3221181, "author_name": "greySnow", "author_url": "", "post_date": "2025-06-10T14:13:38.803000", "content": "", "votes": 3, "replies": [ { "id": 3221205, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2025-06-10T15:09:20.017000", "content": ":D .. Fingers crossed for Bronze
", "votes": 0, "replies": [ { "id": 3221255, "author_name": "greySnow", "author_url": "", "post_date": "2025-06-10T16:37:13.640000", "content": "Winning bronze in Kaggle: the art of being the first to copy and submit notebooks 🤣
", "votes": 6, "replies": [] }, { "id": 3221505, "author_name": "", "author_url": "", "post_date": "2025-06-11T05:29:40.017000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3221218, "author_name": "", "author_url": "", "post_date": "2025-06-10T15:27:13.803000", "content": "", "votes": 0, "replies": [] }, { "id": 3221515, "author_name": "", "author_url": "", "post_date": "2025-06-11T05:41:20.970000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3221755, "author_name": "Truth Seeker", "author_url": "", "post_date": "2025-06-11T12:05:32.587000", "content": "I always wonder, what is the most dramatic public/private ranking jumps observed? For example, if one ranks 1573 on public LB and 102 on private LB, the jump would be +1471.
\nCan some Kagglers share what're the most extreme ranking jumps (both positive and negative) they ever observed?
\nThis is called shakeup, right? Sorry I'm relatively new to Kaggling.
", "votes": 0, "replies": [ { "id": 3221766, "author_name": "greySnow", "author_url": "", "post_date": "2025-06-11T12:14:24.387000", "content": "There are competitions that are pure lottery, with jumps in the thousands. More than a few. I think I saw 4K+ for first place sometime?
\nBut this competition will have very small shake, almost certainly.
https://www.kaggle.com/competitions/blood-vessel-segmentation/leaderboard
\nhttps://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/475657
This is absolutely the most unbelievable ghost story. This player had terrible scores on the public leaderboard and seemingly threw in the towel, but somehow their solution ended up securing second place!
", "votes": 3, "replies": [ { "id": 3221986, "author_name": "", "author_url": "", "post_date": "2025-06-11T17:08:42.250000", "content": "Go check out AIMO-2 and AIMO-1
\nPeople from top research centers across the world (like ETH Zurich) who ruled the public leaderboard for months had reversed polarity on the private set. Happened with many of the top 10-20. And out of nowhere some people with barely 20-30 submissions popped off to the top places.
In AIMO the very top were stable. If you were really good you win, it was just really really hard to be better than the general noise range.
\nYou want to see real lottery? Check this or this.
\nOh and BELKA was absolutely ridiculous, with 3 out of top 5 novices with 10 or less submissions in the competition.
\nBut we usually can spot lottery competitions long before the shake up, and the opposite is also true. This one, for example, will NOT be lottery.
Even though this competition will not be lottery, surely the countless public 28.2 contestants can't all rank better than all the public 28.3+ contestants in private ranking, right?
\nI wonder how many public 28.3+ contestants would actually beat (at least half of) the public 28.2 contestants in private ranking.
\nAny educated clue?
", "votes": 0, "replies": [] }, { "id": 3222075, "author_name": "greySnow", "author_url": "", "post_date": "2025-06-11T19:16:12.723000", "content": "Sure. A lot of competitions are very stable at the top and shaky down the line.
\nThis is already not due to 'lottery' of conpetition, but because it's a hard one so most people can't get good and well validated solutions that will separate them from the noise of the public (in this competition-28.2 is noise range since it's just copy public notebook and maybe change a few parameters).
\nAs long as top solutions are not lottery, its good.
I think it's because we copy the notebook published in Code, run it, and create submission.csv
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3221181": "[Welcome to Kaggle 🤣](https://www.kaggle.com/code/overvalueawareness/caformer-full-resolution-improved)", "3221177": "Hi everyone!\nI wanted to ask a question: Why are there so many people with a score of 28.2?", "3221755": "I always wonder, what is the most dramatic public/private ranking jumps observed? For example, if one ranks 1573 on public LB and 102 on private LB, the jump would be +1471.\n\nCan some Kagglers share what're the most extreme ranking jumps (both positive and negative) they ever observed?\n\nThis is called shakeup, right? Sorry I'm relatively new to Kaggling.", "3221744": "I think it's because we copy the notebook published in Code, run it, and create submission.csv" } }