How do you guys cope up with leaderboard being faster and saturated than your progress? Don't you all lose the motivation to even try when all that people would do is public the top notebooks.
", "messages": [ { "id": 2584785, "postDate": "2024-01-03T07:34:09.963Z", "content": "how to lose a competition:
\nThis is so true that it hurts😅
", "rawMarkdown": "This is so true that it hurts😅 ", "votes": 1 }, { "id": 2584819, "postDate": "2024-01-03T07:48:42.843Z", "content": "😭Sigh…..
", "rawMarkdown": "😭Sigh.....", "replies": [ { "id": 2584854, "postDate": "2024-01-03T08:04:44.677Z", "content": "but you can avoid the last four by believe in yourself and try harder
", "rawMarkdown": "but you can avoid the last four by believe in yourself and try harder", "votes": 3 } ] } ] }, { "id": 2582099, "postDate": "2024-01-01T11:16:18.353Z", "content": "We've fallen from silver to no medal in a period of 2 days 😂, so I figured we'll give our take on this in case you find it helpful:
\nI didn't get your last point
", "rawMarkdown": "I didn't get your last point ", "replies": [ { "id": 2582110, "postDate": "2024-01-01T11:29:51.820Z", "content": "say there are 2 scenarios:
\nI'd prefer to fall into 2., because this way when the competition is over I can learn what would bring me from a 0.85 score to higher scores. Which would include techniques that are more advanced than if I were what I would have learnt if I fall into 1.
", "rawMarkdown": "say there are 2 scenarios:\n1. no one releases anything. I got 0.8 LB score, and lose\n2. someone releases something, I got 0.85 LB score because I took some part of the code, and lose anyway\n\nI'd prefer to fall into 2., because this way when the competition is over I can learn what would bring me from a 0.85 score to higher scores. Which would include techniques that are more advanced than if I were what I would have learnt if I fall into 1.", "votes": 1 }, { "id": 2582146, "postDate": "2024-01-01T12:09:23.153Z", "content": "In my opinion, a month before the competition deadline, high scoring notebooks do more good than harm. Each of us have an opportunity to learn something new and collectively build on the best ideas shared on the forums.
\nHigh scoring notebooks, a week before the competition deadline, hmm, might not be a good idea but that debate is for another day 😀
", "rawMarkdown": "In my opinion, a month before the competition deadline, high scoring notebooks do more good than harm. Each of us have an opportunity to learn something new and collectively build on the best ideas shared on the forums.\n\nHigh scoring notebooks, a week before the competition deadline, hmm, might not be a good idea but that debate is for another day 😀" }, { "id": 2584717, "postDate": "2024-01-03T06:22:59.393Z", "content": "i get it now
", "rawMarkdown": "i get it now" } ] } ] }, { "id": 2581919, "postDate": "2024-01-01T08:23:41.663Z", "content": "The best thing to do is see what can be learned from public notebooks or discussions on training models then try to develop your own work and ideas. Inference notebooks can be useful to get working submissions or ideas for TTA, etc.
\nKeep in mind there is a public LB and a private LB. The private LB is all that matters at the end of the competition. So sometimes it is best to focus on that and not worry too much about public LB positions. Sometimes there are big shakeups!
", "rawMarkdown": "The best thing to do is see what can be learned from public notebooks or discussions on training models then try to develop your own work and ideas. Inference notebooks can be useful to get working submissions or ideas for TTA, etc. \n\nKeep in mind there is a public LB and a private LB. The private LB is all that matters at the end of the competition. So sometimes it is best to focus on that and not worry too much about public LB positions. Sometimes there are big shakeups!", "votes": 5, "replies": [ { "id": 2582108, "postDate": "2024-01-01T11:22:56.630Z", "content": "I kinda lose it when hardwork is beaten by plagiarism
", "rawMarkdown": "I kinda lose it when hardwork is beaten by plagiarism " }, { "id": 2582133, "postDate": "2024-01-01T11:55:51.503Z", "content": "Unfortunately, the public notebooks are not quite compatible with my approach, and they have not provided any improvement to my work, only added pressure.
", "rawMarkdown": "Unfortunately, the public notebooks are not quite compatible with my approach, and they have not provided any improvement to my work, only added pressure.\n\n\n\n\n\n", "votes": 1 } ] }, { "id": 2581910, "postDate": "2024-01-01T08:10:06.713Z", "content": "How do you guys cope up with leaderboard being faster and saturated than your progress? Don't you all lose the motivation to even try when all that people would do is public the top notebooks.
", "rawMarkdown": "How do you guys cope up with leaderboard being faster and saturated than your progress? Don't you all lose the motivation to even try when all that people would do is public the top notebooks. ", "votes": 6 }, { "id": 2582191, "postDate": "2024-01-01T12:47:23.677Z", "content": "I'm still tryng to understand them since some training choices translated to my work make my results worst. It's hard to belive simple hyperparameter tunning gives such good results. I still have my doubts. We'll see.
", "rawMarkdown": "I'm still tryng to understand them since some training choices translated to my work make my results worst. It's hard to belive simple hyperparameter tunning gives such good results. I still have my doubts. We'll see.", "votes": 1, "replies": [ { "id": 2584718, "postDate": "2024-01-03T06:24:01.257Z", "content": "exactly. But again, i feel that i might just be inexperienced yet to realise that hyperparamters are what really matter
", "rawMarkdown": "exactly. But again, i feel that i might just be inexperienced yet to realise that hyperparamters are what really matter" } ] }, { "id": 2584759, "postDate": "2024-01-03T07:12:46.187Z", "content": "the private test data is of different resolution.
\nif you just use public notebook as it is, you will definitely fall off at private leader board
Yes
\nI realised this .
\nI too don't like to use the publicly available submissions
how to lose a competition:
\nThis is so true that it hurts😅
", "votes": 1, "replies": [] }, { "id": 2584819, "author_name": "Huang Jin Feng", "author_url": "", "post_date": "2024-01-03T07:48:42.843000", "content": "😭Sigh…..
", "votes": 0, "replies": [ { "id": 2584854, "author_name": "hengck23", "author_url": "", "post_date": "2024-01-03T08:04:44.677000", "content": "but you can avoid the last four by believe in yourself and try harder
", "votes": 3, "replies": [] } ] } ] }, { "id": 2582099, "author_name": "Sumo", "author_url": "", "post_date": "2024-01-01T11:16:18.353000", "content": "We've fallen from silver to no medal in a period of 2 days 😂, so I figured we'll give our take on this in case you find it helpful:
\nI didn't get your last point
", "votes": 0, "replies": [ { "id": 2582110, "author_name": "Sumo", "author_url": "", "post_date": "2024-01-01T11:29:51.820000", "content": "say there are 2 scenarios:
\nI'd prefer to fall into 2., because this way when the competition is over I can learn what would bring me from a 0.85 score to higher scores. Which would include techniques that are more advanced than if I were what I would have learnt if I fall into 1.
", "votes": 1, "replies": [] }, { "id": 2582146, "author_name": "binga", "author_url": "", "post_date": "2024-01-01T12:09:23.153000", "content": "In my opinion, a month before the competition deadline, high scoring notebooks do more good than harm. Each of us have an opportunity to learn something new and collectively build on the best ideas shared on the forums.
\nHigh scoring notebooks, a week before the competition deadline, hmm, might not be a good idea but that debate is for another day 😀
", "votes": 0, "replies": [] }, { "id": 2584717, "author_name": "Bhavesh Jain", "author_url": "", "post_date": "2024-01-03T06:22:59.393000", "content": "i get it now
", "votes": 0, "replies": [] } ] } ] }, { "id": 2581919, "author_name": "something4kag", "author_url": "", "post_date": "2024-01-01T08:23:41.663000", "content": "The best thing to do is see what can be learned from public notebooks or discussions on training models then try to develop your own work and ideas. Inference notebooks can be useful to get working submissions or ideas for TTA, etc.
\nKeep in mind there is a public LB and a private LB. The private LB is all that matters at the end of the competition. So sometimes it is best to focus on that and not worry too much about public LB positions. Sometimes there are big shakeups!
", "votes": 5, "replies": [ { "id": 2582108, "author_name": "Bhavesh Jain", "author_url": "", "post_date": "2024-01-01T11:22:56.630000", "content": "I kinda lose it when hardwork is beaten by plagiarism
", "votes": 0, "replies": [] }, { "id": 2582133, "author_name": "Huang Jin Feng", "author_url": "", "post_date": "2024-01-01T11:55:51.503000", "content": "Unfortunately, the public notebooks are not quite compatible with my approach, and they have not provided any improvement to my work, only added pressure.
", "votes": 1, "replies": [] } ] }, { "id": 2582191, "author_name": "Ángel Jacinto Sánchez Ruiz", "author_url": "", "post_date": "2024-01-01T12:47:23.677000", "content": "I'm still tryng to understand them since some training choices translated to my work make my results worst. It's hard to belive simple hyperparameter tunning gives such good results. I still have my doubts. We'll see.
", "votes": 1, "replies": [ { "id": 2584718, "author_name": "Bhavesh Jain", "author_url": "", "post_date": "2024-01-03T06:24:01.257000", "content": "exactly. But again, i feel that i might just be inexperienced yet to realise that hyperparamters are what really matter
", "votes": 0, "replies": [] } ] }, { "id": 2584759, "author_name": "hengck23", "author_url": "", "post_date": "2024-01-03T07:12:46.187000", "content": "the private test data is of different resolution.
\nif you just use public notebook as it is, you will definitely fall off at private leader board
Yes
\nI realised this .
\nI too don't like to use the publicly available submissions