This competition is fun to watch. Every Kaggle competition during the last week has all teams improve their score a lot. I'm curious to guess what gold medal will be in this competition. It looks like public LB gold medal will be under public LB<=12.0! Good luck everyone!
", "messages": [ { "id": 3232577, "postDate": "2025-06-26T00:09:17.683Z", "content": "This competition is fun to watch. Every Kaggle competition during the last week has all teams improve their score a lot. I'm curious to guess what gold medal will be in this competition. It looks like public LB gold medal will be under public LB<=12.0! Good luck everyone!
", "rawMarkdown": "This competition is fun to watch. Every Kaggle competition during the last week has all teams improve their score a lot. I'm curious to guess what gold medal will be in this competition. It looks like public LB gold medal will be under public LB<=12.0! Good luck everyone!", "votes": 24 }, { "id": 3232624, "postDate": "2025-06-26T02:46:37.273Z", "content": "different lb score range probably ,means slightly diffrent methods.
\ne.g. below 10 : method = …
\nfrom score=s1 to s2 : method = …
my belief is that smiliar score refers to somehow smiliar method. so if the teams uses smiliar method, then gold will be under public lb 12
", "rawMarkdown": "different lb score range probably ,means slightly diffrent methods.\ne.g. below 10 : method = ...\nfrom score=s1 to s2 : method = ...\n\nmy belief is that smiliar score refers to somehow smiliar method. so if the teams uses smiliar method, then gold will be under public lb 12", "votes": 3 }, { "id": 3232870, "postDate": "2025-06-26T09:21:27.747Z", "content": "Seeing the other teams scores improve every minute is less fun when you are a participant :D
\nJust kidding, the finale will be great.
", "rawMarkdown": "Seeing the other teams scores improve every minute is less fun when you are a participant :D\n\nJust kidding, the finale will be great.", "votes": 4 }, { "id": 3237133, "postDate": "2025-07-01T00:08:22.013Z", "content": "Congratulations everyone! Looks like LB=12.x or less was required for Gold!
", "rawMarkdown": "Congratulations everyone! Looks like LB=12.x or less was required for Gold!", "votes": 1, "replies": [ { "id": 3237135, "postDate": "2025-07-01T00:09:18.517Z", "content": "so you are correct! top kagglers are top predictors!
\nalso gold <= rank12
Yes it's amazing.
\n'Never underestimate Kagglers'.
\nI must said I was really surprised.
\nI expected to take solo price easily.
\n(Well I did take it but very not easily and it was very close to not to).
\nThis competition was really something else.
haha, I've been in 90+ competitions. One thing I learned is that the progress of Kagglers in the last 7 days is always more than I expect. (So when making predictions, we need to predict the unexpected 😀 )
", "rawMarkdown": "haha, I've been in 90+ competitions. One thing I learned is that the progress of Kagglers in the last 7 days is **always** more than I expect. (So when making predictions, we need to predict the unexpected 😀 )", "votes": 2 }, { "id": 3237169, "postDate": "2025-07-01T00:30:02.263Z", "content": "Congratulations Hengck23 and team, and Congratulations GreySnow achieving Gold Medal!
", "rawMarkdown": "Congratulations Hengck23 and team, and Congratulations GreySnow achieving Gold Medal!", "votes": 1 } ] }, { "id": 3233835, "postDate": "2025-06-27T08:15:23.010Z", "content": "Seems like all the forteen’s are fitting for the last prize slot. Tough fight!
", "rawMarkdown": "Seems like all the forteen’s are fitting for the last prize slot. Tough fight!", "votes": 1 }, { "id": 3232938, "postDate": "2025-06-26T10:26:18.137Z", "content": "Everyone is pushing so hard—it looks like the gold line will most likely be 16+ now. If 15 ends up not being enough for gold, that would be absolutely crazy.
", "rawMarkdown": "Everyone is pushing so hard—it looks like the gold line will most likely be 16+ now. If 15 ends up not being enough for gold, that would be absolutely crazy.", "votes": 2, "replies": [ { "id": 3232960, "postDate": "2025-06-26T10:56:58.680Z", "content": "Assuming private test distribution is similar to public test distribution, I think gold will be below 14. Could be way less even, hard to tell now. reason is how fast teams ar e moving, simple extrapolation.
\nWhat I find more interesting is estimating the winner score. I wonder if it will be above or below 5.
", "rawMarkdown": "Assuming private test distribution is similar to public test distribution, I think gold will be below 14. Could be way less even, hard to tell now. reason is how fast teams ar e moving, simple extrapolation.\n\nWhat I find more interesting is estimating the winner score. I wonder if it will be above or below 5.", "votes": 2, "replies": [ { "id": 3232966, "postDate": "2025-06-26T11:02:21.473Z", "content": "It depends. The lower u go for public, the higher is the risk to overfitting to private.. Assuming one can control the amount of fitting, it is a matter of choice
", "rawMarkdown": "It depends. The lower u go for public, the higher is the risk to overfitting to private.. Assuming one can control the amount of fitting, it is a matter of choice\n\n\n", "votes": 1 }, { "id": 3232976, "postDate": "2025-06-26T11:11:34.360Z", "content": "You guys always make me nervous! Since training is so time-consuming, most teams have probably already done their model fusion in advance. Adding a few more in the final days likely won’t make a huge difference—I still think 15 has a pretty solid shot at gold.
", "rawMarkdown": "You guys always make me nervous! Since training is so time-consuming, most teams have probably already done their model fusion in advance. Adding a few more in the final days likely won’t make a huge difference—I still think 15 has a pretty solid shot at gold." }, { "id": 3232984, "postDate": "2025-06-26T11:27:50.180Z", "content": "< 14-15 seems reasonable.
\nAlso 2/3/4 place teams progress much slower once we hit sub-12 (even tho we all submit daily and I presume these are all good/valid submissions). Previously we had much larger jumps like >1 MAE, now we have by an order of magnitude smaller LB jumps. So IMO 12 is the lower bound.
", "rawMarkdown": "< 14-15 seems reasonable.\n\nAlso 2/3/4 place teams progress much slower once we hit sub-12 (even tho we all submit daily and I presume these are all good/valid submissions). Previously we had much larger jumps like >1 MAE, now we have by an order of magnitude smaller LB jumps. So IMO 12 is the lower bound.", "votes": 1 }, { "id": 3233000, "postDate": "2025-06-26T11:44:49.860Z", "content": "Teacher Huang, are you still planning to continue hiding your score? I think it's very risky to keep doing that.
", "rawMarkdown": "Teacher Huang, are you still planning to continue hiding your score? I think it's very risky to keep doing that.", "votes": 1 }, { "id": 3233003, "postDate": "2025-06-26T11:49:22.320Z", "content": "\n\nAlso 2/3/4 place teams progress much slower once we hit sub-12 (even tho we all submit daily and I presume these are all good/valid submissions). Previously we had much larger jumps like >1 MAE, now we have by an order of magnitude smaller LB jumps. So IMO 12 is the lower bound.
\n
Can confirm. It's hard.
", "rawMarkdown": ">Also 2/3/4 place teams progress much slower once we hit sub-12 (even tho we all submit daily and I presume these are all good/valid submissions). Previously we had much larger jumps like >1 MAE, now we have by an order of magnitude smaller LB jumps. So IMO 12 is the lower bound. \n\nCan confirm. It's hard. ", "votes": 2 }, { "id": 3233005, "postDate": "2025-06-26T11:53:43Z", "content": "one can make a table:
\nfrom 14 to 13 : number of days taken from monitoring lb
\nfrom 13 to 12 : number of days taken
\n….
then can estimate lb from number of remaining days
", "rawMarkdown": "one can make a table:\nfrom 14 to 13 : number of days taken from monitoring lb\nfrom 13 to 12 : number of days taken \n....\n\nthen can estimate lb from number of remaining days", "votes": 1 }, { "id": 3233006, "postDate": "2025-06-26T11:54:06.573Z", "content": "\n\nTeacher Huang, are you still planning to continue hiding your score? I think it's very risky to keep doing that.
\n
Waiting to see @hydantess score is always exciting.
\nIt's part of the fun.
I think he is gigachad enough to hide it till the last day and then 'suddenly' appear in the top-5
", "rawMarkdown": "I think he is gigachad enough to hide it till the last day and then 'suddenly' appear in the top-5", "votes": 1 }, { "id": 3233009, "postDate": "2025-06-26T12:01:19.590Z", "content": "I didn't mean to hide it. I haven't written the submission code yet, and based on the CV gap score, it's probably around 14-15. The risk is indeed very high. During the previous CIBMTR competition, I wrote the submission notebook on the last day, encountered a bug, and only managed to submit it successfully two hours before the competition ended.
", "rawMarkdown": "I didn't mean to hide it. I haven't written the submission code yet, and based on the CV gap score, it's probably around 14-15. The risk is indeed very high. During the previous CIBMTR competition, I wrote the submission notebook on the last day, encountered a bug, and only managed to submit it successfully two hours before the competition ended.", "votes": 4 }, { "id": 3233010, "postDate": "2025-06-26T12:02:29.580Z", "content": "kaggle should make new types of competition:
\nAnd you got gold! you are too lazy in submitting 🤣
", "rawMarkdown": "And you got gold! you are too lazy in submitting 🤣" }, { "id": 3233014, "postDate": "2025-06-26T12:05:14.840Z", "content": "\n\nonly total 5 submission in public LB in 3 months
\n
\n no LB at all (we are now in blind chess)
There is no need to create new format that @hydantess is sure to win each time.
\nHe already win enough as it is 🤣
11.x was a huge wall for me in my validation, so maybe it make sense that it slows down below 12
\nThe wall at 8 seems to be even harder…
Really excited to read your solution @harshitsheoran I have no idea how to get from 11 to 8, it's a huge gap
", "rawMarkdown": "Really excited to read your solution @harshitsheoran I have no idea how to get from 11 to 8, it's a huge gap" }, { "id": 3233047, "postDate": "2025-06-26T12:47:24.133Z", "content": "I am really excited to read a lot of solutions in this competition including yours, I believe that better approaches than mine could have been crafted, not all my decisions may have come through. After failing to find anything that breaks past 8.3 in last few days, at this point I will just pray that my score scales a bit more with more compute in the last 3-4 days of training.
", "rawMarkdown": "I am really excited to read a lot of solutions in this competition including yours, I believe that better approaches than mine could have been crafted, not all my decisions may have come through. After failing to find anything that breaks past 8.3 in last few days, at this point I will just pray that my score scales a bit more with more compute in the last 3-4 days of training.", "votes": 1 }, { "id": 3233247, "postDate": "2025-06-26T15:55:30.743Z", "content": "Before we reach 12. I am always happy , improving at lease 1 MAE every day.
\nBut after that, everything changed….. Exaustedly, but hard to improve. 🫠
", "rawMarkdown": "Before we reach 12. I am always happy , improving at lease 1 MAE every day. \n\nBut after that, everything changed..... Exaustedly, but hard to improve. 🫠", "votes": 3 }, { "id": 3233286, "postDate": "2025-06-26T16:42:15.827Z", "content": "Band of results from the leaderboard is now very obvious:
\nBelow 10
\nAbout 11
\nAbout 14
\nAbout 18
\n…
\nAbout 28
@hengck23 Maybe it's a bit unfair to you as you spent effort doing some amazing work for the RNA folding competition, and that left only one month to work on this competition.
", "rawMarkdown": "@hengck23 Maybe it's a bit unfair to you as you spent effort doing some amazing work for the RNA folding competition, and that left only one month to work on this competition." }, { "id": 3233355, "postDate": "2025-06-26T17:50:33.637Z", "content": "I don’t think so. I won’t use the word “unfair”. A better word is “maybe one seems to be at disadvantage to join the competition late and has less time to spend on it …”
\nBut this is case by case. For your information, when Microsoft built human pose estimator for xbox, the computer vision team only took one week to show management a working prototype. They had to build from scratch as depth camera is a new product.
\nWorking in the industrial work before, I would say rush work is common. Normally I was given on week to show results, which will decide the project to go on or go off.
\nI would say one week for top 50 and a month to work to the top is a comfortable time that we must adapted to.
", "rawMarkdown": "I don’t think so. I won’t use the word “unfair”. A better word is “maybe one seems to be at disadvantage to join the competition late and has less time to spend on it …”\n\nBut this is case by case. For your information, when Microsoft built human pose estimator for xbox, the computer vision team only took one week to show management a working prototype. They had to build from scratch as depth camera is a new product.\n\nWorking in the industrial work before, I would say rush work is common. Normally I was given on week to show results, which will decide the project to go on or go off. \n\nI would say one week for top 50 and a month to work to the top is a comfortable time that we must adapted to.", "votes": 3 } ] } ] }, { "id": 3232596, "postDate": "2025-06-26T01:44:35.113Z", "content": "I wanted to join this comp so bad but my wallet was not as excited so we have been pretty stagnant since the public model came out. Our strategy relies on needing the model itself to improve score sadly.
", "rawMarkdown": "I wanted to join this comp so bad but my wallet was not as excited so we have been pretty stagnant since the public model came out. Our strategy relies on needing the model itself to improve score sadly. ", "votes": 1, "replies": [ { "id": 3234290, "postDate": "2025-06-27T17:39:47.143Z", "content": "help me to learn more about competitions on kaggle please
", "rawMarkdown": "help me to learn more about competitions on kaggle please\n" } ] }, { "id": 3236249, "postDate": "2025-06-30T06:46:50.803Z", "content": "This competition has been really fun to follow! As always, the final week brings a lot of action and leaderboard shakeups. I'm curious to see where the gold medal cutoff lands — with all the improvements, it looks like it might end up below 12.0 on the public LB!
", "rawMarkdown": "This competition has been really fun to follow! As always, the final week brings a lot of action and leaderboard shakeups. I'm curious to see where the gold medal cutoff lands — with all the improvements, it looks like it might end up below 12.0 on the public LB! \n", "votes": -2 }, { "id": 3237560, "postDate": "2025-07-01T07:02:51.423Z", "content": "How can I learn to joinn competitions on Kaggle?
", "rawMarkdown": "How can I learn to joinn competitions on Kaggle?", "replies": [ { "id": 3300886, "postDate": "2025-10-11T22:37:34.943Z", "content": "Tell me too , I wanna join competitions
", "rawMarkdown": "Tell me too , I wanna join competitions\n" } ] }, { "id": 3237058, "postDate": "2025-06-30T21:36:24.383Z", "content": "a couple of one-shot kagglers. maybe also zero-shot ones
", "rawMarkdown": " a couple of one-shot kagglers. maybe also zero-shot ones", "replies": [ { "id": 3237078, "postDate": "2025-06-30T22:24:52.557Z", "content": "I guess they did a number of experiments before one submission.
", "rawMarkdown": "I guess they did a number of experiments before one submission. " }, { "id": 3237106, "postDate": "2025-06-30T23:26:22.743Z", "content": "one-shot kagglers : one submission
\nzero-shot : no submission
I'm surprised 14.4 isn't enough to get gold medal.
", "rawMarkdown": "I'm surprised 14.4 isn't enough to get gold medal.", "replies": [ { "id": 3236263, "postDate": "2025-06-30T07:28:42.320Z", "content": "13.x won't be enough to get gold IMHO. I need to move :D
\nNice jump BTW.
", "rawMarkdown": "13.x won't be enough to get gold IMHO. I need to move :D\n\nNice jump BTW.", "replies": [ { "id": 3236276, "postDate": "2025-06-30T07:56:30.377Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3236987, "postDate": "2025-06-30T19:46:19.867Z", "content": "\n\n13.x won't be enough to get gold IMHO.
\n
Seems I'll be right (if private LB is similar to public LB)
", "rawMarkdown": "> 13.x won't be enough to get gold IMHO.\n\nSeems I'll be right (if private LB is similar to public LB)", "votes": 1 } ] }, { "id": 3236277, "postDate": "2025-06-30T07:56:41.193Z", "content": "Me too, I’ve been staying up for a long time watching over the training without sleep, and now I feel like I’m basically done for.
", "rawMarkdown": "Me too, I’ve been staying up for a long time watching over the training without sleep, and now I feel like I’m basically done for.", "replies": [ { "id": 3236307, "postDate": "2025-06-30T08:34:34.227Z", "content": "me still trying same github code, etc, don't think can finish training these but just to get as much exposure as possible …
", "rawMarkdown": "me still trying same github code, etc, don't think can finish training these but just to get as much exposure as possible ... ", "votes": 1 }, { "id": 3236315, "postDate": "2025-06-30T08:42:27.403Z", "content": "You've done a really great job. I realized the right approach too late
", "rawMarkdown": "You've done a really great job. I realized the right approach too late" } ] }, { "id": 3236282, "postDate": "2025-06-30T08:03:18.810Z", "content": "this is a shakeup of +/- one if generalisation is not well controlled.
\nso 13.x has chance as for leaderboard now (don't konw after 12 hours later)
Host said the data is the same distribution as train, so I believe shake will be more in the range of 0.1.
\nUnless host lied…(then it can be funny 🤣)
A realistic situation would be that the family ratios of private and public are different.
", "rawMarkdown": "A realistic situation would be that the family ratios of private and public are different." }, { "id": 3237004, "postDate": "2025-06-30T20:01:53.377Z", "content": "From probing: this is not the case for StyleA, StyleB, and 3 subsets of the other families.
", "rawMarkdown": "From probing: this is not the case for StyleA, StyleB, and 3 subsets of the other families." }, { "id": 3237009, "postDate": "2025-06-30T20:08:52.213Z", "content": "Indeed, I feel stupid to discover it today! 🤣
\nWell, lesson learned.
I am sorry, english is not native for me.
\nAre we referring to how train and test 'method'/'family' distribution is a lot different from each other? (knew this part), or is it something else I am missing?
Yes, we refer to this.
", "rawMarkdown": "Yes, we refer to this." }, { "id": 3237018, "postDate": "2025-06-30T20:20:47.193Z", "content": "For 5 subsets of the families, I was able to probe how often they occur in the total test set and in the public test set. For these 5 subsets, the ratios are nearly equal. So the private test set seems to have the same distribution of families as the public test set (and both are different from the train distribution).
", "rawMarkdown": "For 5 subsets of the families, I was able to probe how often they occur in the total test set and in the public test set. For these 5 subsets, the ratios are nearly equal. So the private test set seems to have the same distribution of families as the public test set (and both are different from the train distribution).", "votes": 1 }, { "id": 3237019, "postDate": "2025-06-30T20:22:44.497Z", "content": "Also @harshitsheoran , I will not be making any more submissions 🙂
", "rawMarkdown": "Also @harshitsheoran , I will not be making any more submissions 🙂", "votes": 1 }, { "id": 3237028, "postDate": "2025-06-30T20:32:10.690Z", "content": "@jeroencottaar
\nThank you so much for informing me, it really relaxes my anxiety buildup
Please refrain from sharing before competition end.
", "rawMarkdown": "Please refrain from sharing before competition end." } ] }, { "id": 3236304, "postDate": "2025-06-30T08:31:31.867Z", "content": "hyd have not submitted yet also…
", "rawMarkdown": "hyd have not submitted yet also...", "replies": [ { "id": 3236309, "postDate": "2025-06-30T08:35:32.280Z", "content": "NOW I know what hyd is aiming at!
\nhttps://x.com/hydantess1993/status/1912115912930427233?s=46&t=l_EidCh6K02wivt3ARL8Fw
My cv is 8.1, i guess lb score is around 12. Don't worry!
", "rawMarkdown": "My cv is 8.1, i guess lb score is around 12. Don't worry!", "votes": 1 }, { "id": 3236320, "postDate": "2025-06-30T08:50:07.497Z", "content": "No, cv follow LB much more closely.
\nMaybe you junp to 2nd.
\n1st not likely.
1 submission 2nd place is still awesome!
", "rawMarkdown": "1 submission 2nd place is still awesome!", "votes": 1 }, { "id": 3236333, "postDate": "2025-06-30T09:00:35.007Z", "content": "A single submission rank 1 would be so so cool that I want you to win this :D
", "rawMarkdown": "A single submission rank 1 would be so so cool that I want you to win this :D" }, { "id": 3236335, "postDate": "2025-06-30T09:03:15.383Z", "content": "Really? I am using Bartley's fold, maybe not closely~
", "rawMarkdown": "Really? I am using Bartley's fold, maybe not closely~ \n" }, { "id": 3236340, "postDate": "2025-06-30T09:07:23.220Z", "content": "My CV is not much far from yours (a bit better), although I am not using Bartley's fold, my fold set is 30k samples and is a harder set IMO
", "rawMarkdown": "My CV is not much far from yours (a bit better), although I am not using Bartley's fold, my fold set is 30k samples and is a harder set IMO" }, { "id": 3236412, "postDate": "2025-06-30T09:25:37.467Z", "content": "Yeah, my 8.1 almost lasted for a week, very hard to improve any further, You deserve to win~ I am satisfied with top 5 finish(if so)!
", "rawMarkdown": "Yeah, my 8.1 almost lasted for a week, very hard to improve any further, You deserve to win~ I am satisfied with top 5 finish(if so)!" }, { "id": 3237129, "postDate": "2025-07-01T00:05:47.580Z", "content": "congrats to @tascj0 @christofhenkel @hydantess 😃😃😃
", "rawMarkdown": "congrats to @tascj0 @christofhenkel @hydantess 😃😃😃", "votes": 2 }, { "id": 3237752, "postDate": "2025-07-01T09:27:21.420Z", "content": "Congrats on strong single sub finish. Looks like your CV was closer to LB than Bartley reported. I think its because of early stopping he did, i.e. picking best CV epochs. That overfits to CV.
", "rawMarkdown": "Congrats on strong single sub finish. Looks like your CV was closer to LB than Bartley reported. I think its because of early stopping he did, i.e. picking best CV epochs. That overfits to CV." }, { "id": 3237779, "postDate": "2025-07-01T09:56:45.360Z", "content": "Yes, the score was a bit higher than I expected. I'm not very good at computer vision competitions, and the probability of one submission win is too low. I'll try other types of competitions in the future.
", "rawMarkdown": "Yes, the score was a bit higher than I expected. I'm not very good at computer vision competitions, and the probability of one submission win is too low. I'll try other types of competitions in the future.", "votes": 1 } ] }, { "id": 3236306, "postDate": "2025-06-30T08:33:17.873Z", "content": "13.x is so far away.
\nOur team can only pray for a shakeup.
The greatest hope for some, the greatest fear for others 😛
", "rawMarkdown": "The greatest hope for some, the greatest fear for others 😛", "votes": 3 } ] } ] }, { "id": 3234225, "postDate": "2025-06-27T16:25:45.480Z", "content": "I hope you describe in detail what you did to achieve such a result
\nI find it really interesting 😉
Kaggle should make a new competition such that winner has lb score closest to a reference, eg reference= 1.2345678
\nwinner are those that can really manipulated non visible data+model and get any score they want.
\nKing of fitting
How can I learn to join competitions on Kaggle?
", "rawMarkdown": "How can I learn to join competitions on Kaggle?", "isDeleted": true, "replies": [ { "id": 3236264, "postDate": "2025-06-30T07:29:20.803Z", "content": "https://www.kaggle.com/competitions/waveform-inversion/discussion/572279
", "rawMarkdown": "https://www.kaggle.com/competitions/waveform-inversion/discussion/572279" } ] } ], "comments": [ { "id": 3232624, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-26T02:46:37.273000", "content": "different lb score range probably ,means slightly diffrent methods.
\ne.g. below 10 : method = …
\nfrom score=s1 to s2 : method = …
my belief is that smiliar score refers to somehow smiliar method. so if the teams uses smiliar method, then gold will be under public lb 12
", "votes": 3, "replies": [] }, { "id": 3232870, "author_name": "CPMP", "author_url": "", "post_date": "2025-06-26T09:21:27.747000", "content": "Seeing the other teams scores improve every minute is less fun when you are a participant :D
\nJust kidding, the finale will be great.
", "votes": 4, "replies": [] }, { "id": 3237133, "author_name": "Chris Deotte", "author_url": "", "post_date": "2025-07-01T00:08:22.013000", "content": "Congratulations everyone! Looks like LB=12.x or less was required for Gold!
", "votes": 1, "replies": [ { "id": 3237135, "author_name": "hengck23", "author_url": "", "post_date": "2025-07-01T00:09:18.517000", "content": "so you are correct! top kagglers are top predictors!
\nalso gold <= rank12
Yes it's amazing.
\n'Never underestimate Kagglers'.
\nI must said I was really surprised.
\nI expected to take solo price easily.
\n(Well I did take it but very not easily and it was very close to not to).
\nThis competition was really something else.
haha, I've been in 90+ competitions. One thing I learned is that the progress of Kagglers in the last 7 days is always more than I expect. (So when making predictions, we need to predict the unexpected 😀 )
", "votes": 2, "replies": [] }, { "id": 3237169, "author_name": "Chris Deotte", "author_url": "", "post_date": "2025-07-01T00:30:02.263000", "content": "Congratulations Hengck23 and team, and Congratulations GreySnow achieving Gold Medal!
", "votes": 1, "replies": [] } ] }, { "id": 3233835, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-27T08:15:23.010000", "content": "Seems like all the forteen’s are fitting for the last prize slot. Tough fight!
", "votes": 1, "replies": [] }, { "id": 3232938, "author_name": "hyd", "author_url": "", "post_date": "2025-06-26T10:26:18.137000", "content": "Everyone is pushing so hard—it looks like the gold line will most likely be 16+ now. If 15 ends up not being enough for gold, that would be absolutely crazy.
", "votes": 2, "replies": [ { "id": 3232960, "author_name": "CPMP", "author_url": "", "post_date": "2025-06-26T10:56:58.680000", "content": "Assuming private test distribution is similar to public test distribution, I think gold will be below 14. Could be way less even, hard to tell now. reason is how fast teams ar e moving, simple extrapolation.
\nWhat I find more interesting is estimating the winner score. I wonder if it will be above or below 5.
", "votes": 2, "replies": [ { "id": 3232966, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-26T11:02:21.473000", "content": "It depends. The lower u go for public, the higher is the risk to overfitting to private.. Assuming one can control the amount of fitting, it is a matter of choice
", "votes": 1, "replies": [] }, { "id": 3232976, "author_name": "hyd", "author_url": "", "post_date": "2025-06-26T11:11:34.360000", "content": "You guys always make me nervous! Since training is so time-consuming, most teams have probably already done their model fusion in advance. Adding a few more in the final days likely won’t make a huge difference—I still think 15 has a pretty solid shot at gold.
", "votes": 0, "replies": [] }, { "id": 3232984, "author_name": "Yurnero", "author_url": "", "post_date": "2025-06-26T11:27:50.180000", "content": "< 14-15 seems reasonable.
\nAlso 2/3/4 place teams progress much slower once we hit sub-12 (even tho we all submit daily and I presume these are all good/valid submissions). Previously we had much larger jumps like >1 MAE, now we have by an order of magnitude smaller LB jumps. So IMO 12 is the lower bound.
", "votes": 1, "replies": [] }, { "id": 3233000, "author_name": "guo dashuai", "author_url": "", "post_date": "2025-06-26T11:44:49.860000", "content": "Teacher Huang, are you still planning to continue hiding your score? I think it's very risky to keep doing that.
", "votes": 1, "replies": [] }, { "id": 3233003, "author_name": "greySnow", "author_url": "", "post_date": "2025-06-26T11:49:22.320000", "content": "\n\nAlso 2/3/4 place teams progress much slower once we hit sub-12 (even tho we all submit daily and I presume these are all good/valid submissions). Previously we had much larger jumps like >1 MAE, now we have by an order of magnitude smaller LB jumps. So IMO 12 is the lower bound.
\n
Can confirm. It's hard.
", "votes": 2, "replies": [] }, { "id": 3233005, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-26T11:53:43", "content": "one can make a table:
\nfrom 14 to 13 : number of days taken from monitoring lb
\nfrom 13 to 12 : number of days taken
\n….
then can estimate lb from number of remaining days
", "votes": 1, "replies": [] }, { "id": 3233006, "author_name": "greySnow", "author_url": "", "post_date": "2025-06-26T11:54:06.573000", "content": "\n\nTeacher Huang, are you still planning to continue hiding your score? I think it's very risky to keep doing that.
\n
Waiting to see @hydantess score is always exciting.
\nIt's part of the fun.
I think he is gigachad enough to hide it till the last day and then 'suddenly' appear in the top-5
", "votes": 1, "replies": [] }, { "id": 3233009, "author_name": "hyd", "author_url": "", "post_date": "2025-06-26T12:01:19.590000", "content": "I didn't mean to hide it. I haven't written the submission code yet, and based on the CV gap score, it's probably around 14-15. The risk is indeed very high. During the previous CIBMTR competition, I wrote the submission notebook on the last day, encountered a bug, and only managed to submit it successfully two hours before the competition ended.
", "votes": 4, "replies": [] }, { "id": 3233010, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-26T12:02:29.580000", "content": "kaggle should make new types of competition:
\nAnd you got gold! you are too lazy in submitting 🤣
", "votes": 0, "replies": [] }, { "id": 3233014, "author_name": "greySnow", "author_url": "", "post_date": "2025-06-26T12:05:14.840000", "content": "\n\nonly total 5 submission in public LB in 3 months
\n
\n no LB at all (we are now in blind chess)
There is no need to create new format that @hydantess is sure to win each time.
\nHe already win enough as it is 🤣
11.x was a huge wall for me in my validation, so maybe it make sense that it slows down below 12
\nThe wall at 8 seems to be even harder…
Really excited to read your solution @harshitsheoran I have no idea how to get from 11 to 8, it's a huge gap
", "votes": 0, "replies": [] }, { "id": 3233047, "author_name": "Harshit Sheoran", "author_url": "", "post_date": "2025-06-26T12:47:24.133000", "content": "I am really excited to read a lot of solutions in this competition including yours, I believe that better approaches than mine could have been crafted, not all my decisions may have come through. After failing to find anything that breaks past 8.3 in last few days, at this point I will just pray that my score scales a bit more with more compute in the last 3-4 days of training.
", "votes": 1, "replies": [] }, { "id": 3233247, "author_name": "ForcewithMe", "author_url": "", "post_date": "2025-06-26T15:55:30.743000", "content": "Before we reach 12. I am always happy , improving at lease 1 MAE every day.
\nBut after that, everything changed….. Exaustedly, but hard to improve. 🫠
", "votes": 3, "replies": [] }, { "id": 3233286, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-26T16:42:15.827000", "content": "Band of results from the leaderboard is now very obvious:
\nBelow 10
\nAbout 11
\nAbout 14
\nAbout 18
\n…
\nAbout 28
@hengck23 Maybe it's a bit unfair to you as you spent effort doing some amazing work for the RNA folding competition, and that left only one month to work on this competition.
", "votes": 0, "replies": [] }, { "id": 3233355, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-26T17:50:33.637000", "content": "I don’t think so. I won’t use the word “unfair”. A better word is “maybe one seems to be at disadvantage to join the competition late and has less time to spend on it …”
\nBut this is case by case. For your information, when Microsoft built human pose estimator for xbox, the computer vision team only took one week to show management a working prototype. They had to build from scratch as depth camera is a new product.
\nWorking in the industrial work before, I would say rush work is common. Normally I was given on week to show results, which will decide the project to go on or go off.
\nI would say one week for top 50 and a month to work to the top is a comfortable time that we must adapted to.
", "votes": 3, "replies": [] } ] } ] }, { "id": 3232596, "author_name": "Cody_Null", "author_url": "", "post_date": "2025-06-26T01:44:35.113000", "content": "I wanted to join this comp so bad but my wallet was not as excited so we have been pretty stagnant since the public model came out. Our strategy relies on needing the model itself to improve score sadly.
", "votes": 1, "replies": [ { "id": 3234290, "author_name": "Abhishek Awasthi", "author_url": "", "post_date": "2025-06-27T17:39:47.143000", "content": "help me to learn more about competitions on kaggle please
", "votes": 0, "replies": [] } ] }, { "id": 3236249, "author_name": "", "author_url": "", "post_date": "2025-06-30T06:46:50.803000", "content": "This competition has been really fun to follow! As always, the final week brings a lot of action and leaderboard shakeups. I'm curious to see where the gold medal cutoff lands — with all the improvements, it looks like it might end up below 12.0 on the public LB!
", "votes": -2, "replies": [] }, { "id": 3237560, "author_name": "Khushi Yadav", "author_url": "", "post_date": "2025-07-01T07:02:51.423000", "content": "How can I learn to joinn competitions on Kaggle?
", "votes": 0, "replies": [ { "id": 3300886, "author_name": "Abhishek Awasthi", "author_url": "", "post_date": "2025-10-11T22:37:34.943000", "content": "Tell me too , I wanna join competitions
", "votes": 0, "replies": [] } ] }, { "id": 3237058, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-30T21:36:24.383000", "content": "a couple of one-shot kagglers. maybe also zero-shot ones
", "votes": 0, "replies": [ { "id": 3237078, "author_name": "Focus", "author_url": "", "post_date": "2025-06-30T22:24:52.557000", "content": "I guess they did a number of experiments before one submission.
", "votes": 0, "replies": [] }, { "id": 3237106, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-30T23:26:22.743000", "content": "one-shot kagglers : one submission
\nzero-shot : no submission
I'm surprised 14.4 isn't enough to get gold medal.
", "votes": 0, "replies": [ { "id": 3236263, "author_name": "CPMP", "author_url": "", "post_date": "2025-06-30T07:28:42.320000", "content": "13.x won't be enough to get gold IMHO. I need to move :D
\nNice jump BTW.
", "votes": 0, "replies": [ { "id": 3236276, "author_name": "", "author_url": "", "post_date": "2025-06-30T07:56:30.377000", "content": "", "votes": 0, "replies": [] }, { "id": 3236987, "author_name": "CPMP", "author_url": "", "post_date": "2025-06-30T19:46:19.867000", "content": "\n\n13.x won't be enough to get gold IMHO.
\n
Seems I'll be right (if private LB is similar to public LB)
", "votes": 1, "replies": [] } ] }, { "id": 3236277, "author_name": "guo dashuai", "author_url": "", "post_date": "2025-06-30T07:56:41.193000", "content": "Me too, I’ve been staying up for a long time watching over the training without sleep, and now I feel like I’m basically done for.
", "votes": 0, "replies": [ { "id": 3236307, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-30T08:34:34.227000", "content": "me still trying same github code, etc, don't think can finish training these but just to get as much exposure as possible …
", "votes": 1, "replies": [] }, { "id": 3236315, "author_name": "guo dashuai", "author_url": "", "post_date": "2025-06-30T08:42:27.403000", "content": "You've done a really great job. I realized the right approach too late
", "votes": 0, "replies": [] } ] }, { "id": 3236282, "author_name": "hengck23", "author_url": "", "post_date": "2025-06-30T08:03:18.810000", "content": "this is a shakeup of +/- one if generalisation is not well controlled.
\nso 13.x has chance as for leaderboard now (don't konw after 12 hours later)
Host said the data is the same distribution as train, so I believe shake will be more in the range of 0.1.
\nUnless host lied…(then it can be funny 🤣)
A realistic situation would be that the family ratios of private and public are different.
", "votes": 0, "replies": [] }, { "id": 3237004, "author_name": "Jeroen Cottaar", "author_url": "", "post_date": "2025-06-30T20:01:53.377000", "content": "From probing: this is not the case for StyleA, StyleB, and 3 subsets of the other families.
", "votes": 0, "replies": [] }, { "id": 3237009, "author_name": "greySnow", "author_url": "", "post_date": "2025-06-30T20:08:52.213000", "content": "Indeed, I feel stupid to discover it today! 🤣
\nWell, lesson learned.
I am sorry, english is not native for me.
\nAre we referring to how train and test 'method'/'family' distribution is a lot different from each other? (knew this part), or is it something else I am missing?
Yes, we refer to this.
", "votes": 0, "replies": [] }, { "id": 3237018, "author_name": "Jeroen Cottaar", "author_url": "", "post_date": "2025-06-30T20:20:47.193000", "content": "For 5 subsets of the families, I was able to probe how often they occur in the total test set and in the public test set. For these 5 subsets, the ratios are nearly equal. So the private test set seems to have the same distribution of families as the public test set (and both are different from the train distribution).
", "votes": 1, "replies": [] }, { "id": 3237019, "author_name": "Jeroen Cottaar", "author_url": "", "post_date": "2025-06-30T20:22:44.497000", "content": "Also @harshitsheoran , I will not be making any more submissions 🙂
", "votes": 1, "replies": [] }, { "id": 3237028, "author_name": "Harshit Sheoran", "author_url": "", "post_date": "2025-06-30T20:32:10.690000", "content": "@jeroencottaar
\nThank you so much for informing me, it really relaxes my anxiety buildup
Please refrain from sharing before competition end.
", "votes": 0, "replies": [] } ] }, { "id": 3236304, "author_name": "Yurnero", "author_url": "", "post_date": "2025-06-30T08:31:31.867000", "content": "hyd have not submitted yet also…
", "votes": 0, "replies": [ { "id": 3236309, "author_name": "c-number", "author_url": "", "post_date": "2025-06-30T08:35:32.280000", "content": "NOW I know what hyd is aiming at!
\nhttps://x.com/hydantess1993/status/1912115912930427233?s=46&t=l_EidCh6K02wivt3ARL8Fw
My cv is 8.1, i guess lb score is around 12. Don't worry!
", "votes": 1, "replies": [] }, { "id": 3236320, "author_name": "greySnow", "author_url": "", "post_date": "2025-06-30T08:50:07.497000", "content": "No, cv follow LB much more closely.
\nMaybe you junp to 2nd.
\n1st not likely.
1 submission 2nd place is still awesome!
", "votes": 1, "replies": [] }, { "id": 3236333, "author_name": "Harshit Sheoran", "author_url": "", "post_date": "2025-06-30T09:00:35.007000", "content": "A single submission rank 1 would be so so cool that I want you to win this :D
", "votes": 0, "replies": [] }, { "id": 3236335, "author_name": "hyd", "author_url": "", "post_date": "2025-06-30T09:03:15.383000", "content": "Really? I am using Bartley's fold, maybe not closely~
", "votes": 0, "replies": [] }, { "id": 3236340, "author_name": "Harshit Sheoran", "author_url": "", "post_date": "2025-06-30T09:07:23.220000", "content": "My CV is not much far from yours (a bit better), although I am not using Bartley's fold, my fold set is 30k samples and is a harder set IMO
", "votes": 0, "replies": [] }, { "id": 3236412, "author_name": "hyd", "author_url": "", "post_date": "2025-06-30T09:25:37.467000", "content": "Yeah, my 8.1 almost lasted for a week, very hard to improve any further, You deserve to win~ I am satisfied with top 5 finish(if so)!
", "votes": 0, "replies": [] }, { "id": 3237129, "author_name": "hyd", "author_url": "", "post_date": "2025-07-01T00:05:47.580000", "content": "congrats to @tascj0 @christofhenkel @hydantess 😃😃😃
", "votes": 2, "replies": [] }, { "id": 3237752, "author_name": "Dieter", "author_url": "", "post_date": "2025-07-01T09:27:21.420000", "content": "Congrats on strong single sub finish. Looks like your CV was closer to LB than Bartley reported. I think its because of early stopping he did, i.e. picking best CV epochs. That overfits to CV.
", "votes": 0, "replies": [] }, { "id": 3237779, "author_name": "hyd", "author_url": "", "post_date": "2025-07-01T09:56:45.360000", "content": "Yes, the score was a bit higher than I expected. I'm not very good at computer vision competitions, and the probability of one submission win is too low. I'll try other types of competitions in the future.
", "votes": 1, "replies": [] } ] }, { "id": 3236306, "author_name": "c-number", "author_url": "", "post_date": "2025-06-30T08:33:17.873000", "content": "13.x is so far away.
\nOur team can only pray for a shakeup.
The greatest hope for some, the greatest fear for others 😛
", "votes": 3, "replies": [] } ] } ] }, { "id": 3234225, "author_name": "Cat In Glasses", "author_url": "", "post_date": "2025-06-27T16:25:45.480000", "content": "I hope you describe in detail what you did to achieve such a result
\nI find it really interesting 😉
Kaggle should make a new competition such that winner has lb score closest to a reference, eg reference= 1.2345678
\nwinner are those that can really manipulated non visible data+model and get any score they want.
\nKing of fitting
How can I learn to join competitions on Kaggle?
", "votes": 0, "replies": [ { "id": 3236264, "author_name": "CPMP", "author_url": "", "post_date": "2025-06-30T07:29:20.803000", "content": "https://www.kaggle.com/competitions/waveform-inversion/discussion/572279
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "3232577": "This competition is fun to watch. Every Kaggle competition during the last week has all teams improve their score a lot. I'm curious to guess what gold medal will be in this competition. It looks like public LB gold medal will be under public LB<=12.0! Good luck everyone!", "3232624": "different lb score range probably ,means slightly diffrent methods.\ne.g. below 10 : method = ...\nfrom score=s1 to s2 : method = ...\n\nmy belief is that smiliar score refers to somehow smiliar method. so if the teams uses smiliar method, then gold will be under public lb 12", "3232870": "Seeing the other teams scores improve every minute is less fun when you are a participant :D\n\nJust kidding, the finale will be great.", "3237133": "Congratulations everyone! Looks like LB=12.x or less was required for Gold!", "3233835": "Seems like all the forteen’s are fitting for the last prize slot. Tough fight!", "3232938": "Everyone is pushing so hard—it looks like the gold line will most likely be 16+ now. If 15 ends up not being enough for gold, that would be absolutely crazy.", "3232596": "I wanted to join this comp so bad but my wallet was not as excited so we have been pretty stagnant since the public model came out. Our strategy relies on needing the model itself to improve score sadly. ", "3236249": "This competition has been really fun to follow! As always, the final week brings a lot of action and leaderboard shakeups. I'm curious to see where the gold medal cutoff lands — with all the improvements, it looks like it might end up below 12.0 on the public LB! \n", "3237560": "How can I learn to joinn competitions on Kaggle?", "3237058": " a couple of one-shot kagglers. maybe also zero-shot ones", "3236254": "I'm surprised 14.4 isn't enough to get gold medal.", "3234225": "I hope you describe in detail what you did to achieve such a result\nI find it really interesting 😉", "3232947": "Kaggle should make a new competition such that winner has lb score closest to a reference, eg reference= 1.2345678\n\nwinner are those that can really manipulated non visible data+model and get any score they want. \nKing of fitting\n\n", "3234822": "How can I learn to join competitions on Kaggle?" } }