![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2875376%2F0ffdaf8e7346884df03cb45630cedc59%2FScreen%20Shot%202024-07-09%20at%209.07.32%20AM.png?generation=1720496323948863&alt=media\")
Hey there!
\nFirst of all, congratulations to everybody who won and also who didn't! Don't be gutted, we all stuck through! :)
\nThis has to be my favorite competition till date and the first one, I indeed went all in. Here I am, 328 submissions and 99 lb positions drop later, still quite happy!
\nI ensembled multiple models, a few public and most of my own [around 40] for my final submission [0.251 private lb].
\nBut, then there was this one model, my slight variation of Belka 1D CNN public notebook by @ahmedelfazouan ,
\ngave a private score of 0.310.
\nA notebook I ran 2 months ago and evidently forgot! (Shouldn't have! XP)
\nI am sure, I am not alone, there must be a bunch of people who have better private lb, but didn't go onto to select it. I hope, we all learn this skill with time :D
Although, the competition has ended, I thought of still sharing my results, for the ultimate purpose is to contribute a result that could help the community and the hosts to get everything out of this competition.
\nSolution: Used Ah's public notebook, trained the notebook for 50 epochs [Fold 0], and selected the weights of the 28th epoch.
\nPrivate LB: 0.310 and public lb : 0.386
Here's the notebook link : 0.310 private lb
", "messages": [ { "id": 2912727, "postDate": "2024-07-09T03:49:31.080Z", "content": "Hey there!
\nFirst of all, congratulations to everybody who won and also who didn't! Don't be gutted, we all stuck through! :)
\nThis has to be my favorite competition till date and the first one, I indeed went all in. Here I am, 328 submissions and 99 lb positions drop later, still quite happy!
\nI ensembled multiple models, a few public and most of my own [around 40] for my final submission [0.251 private lb].
\nBut, then there was this one model, my slight variation of Belka 1D CNN public notebook by @ahmedelfazouan ,
\ngave a private score of 0.310.
\nA notebook I ran 2 months ago and evidently forgot! (Shouldn't have! XP)
\nI am sure, I am not alone, there must be a bunch of people who have better private lb, but didn't go onto to select it. I hope, we all learn this skill with time :D
Although, the competition has ended, I thought of still sharing my results, for the ultimate purpose is to contribute a result that could help the community and the hosts to get everything out of this competition.
\nSolution: Used Ah's public notebook, trained the notebook for 50 epochs [Fold 0], and selected the weights of the 28th epoch.
\nPrivate LB: 0.310 and public lb : 0.386
Here's the notebook link : 0.310 private lb
", "rawMarkdown": "Hey there! \nFirst of all, congratulations to everybody who won and also who didn't! Don't be gutted, we all stuck through! :)\nThis has to be my favorite competition till date and the first one, I indeed went all in. Here I am, 328 submissions and 99 lb positions drop later, still quite happy!\nI ensembled multiple models, a few public and most of my own [around 40] for my final submission [0.251 private lb]. \nBut, then there was this one model, my slight variation of Belka 1D CNN public notebook by @ahmedelfazouan , \ngave a private score of 0.310. \nA notebook I ran 2 months ago and evidently forgot! (Shouldn't have! XP)\nI am sure, I am not alone, there must be a bunch of people who have better private lb, but didn't go onto to select it. I hope, we all learn this skill with time :D\n\nAlthough, the competition has ended, I thought of still sharing my results, for the ultimate purpose is to contribute a result that could help the community and the hosts to get everything out of this competition. \n\nSolution: Used Ah's [public notebook](https://www.kaggle.com/code/ahmedelfazouan/belka-1dcnn-starter-with-all-data), **trained the notebook for 50 epochs [Fold 0], and selected the weights of the 28th epoch.**\n**Private LB: 0.310 and public lb : 0.386**\n\n\n\nHere's the notebook link : [0.310 private lb](https://www.kaggle.com/ahsuna123/belka-1dcnn-based-on-ah-s-public-notebook)\n", "votes": 12 }, { "id": 2913048, "postDate": "2024-07-09T07:43:27.577Z", "content": "Got the same, luck matters
", "rawMarkdown": "Got the same, luck matters", "votes": 3, "replies": [ { "id": 2913155, "postDate": "2024-07-09T09:37:16.850Z", "content": "Indeed! After seeing a couple of single submissions in the gold zone, I am sold on the luck part! π
", "rawMarkdown": "Indeed! After seeing a couple of single submissions in the gold zone, I am sold on the luck part! π", "votes": 2 } ] }, { "id": 2913217, "postDate": "2024-07-09T10:35:54.383Z", "content": "I enjoyed this Public Score v Private Score dynamic.
\nClearly some folks with 400+ submissions were teaching the test⦠and not generalizing.
", "rawMarkdown": "I enjoyed this Public Score v Private Score dynamic. \n\nClearly some folks with 400+ submissions were teaching the test... and not generalizing.", "votes": -2, "replies": [ { "id": 2913228, "postDate": "2024-07-09T10:46:27Z", "content": "ahaha, btw, the team that used all the available sibmissions received special congratulations for their submission robustness, showing less variance between public and private scores π
", "rawMarkdown": "ahaha, btw, the team that used all the available sibmissions received special congratulations for their submission robustness, showing less variance between public and private scores π\n\n", "votes": 2, "replies": [ { "id": 2913236, "postDate": "2024-07-09T10:56:17.477Z", "content": "Hahaha! Congratulations for another gold! @antoninadolgorukova π₯³
\n@mattmcdonagh I don't think anybody was trying to teach the test thou! :)
\nWe all knew the drop in scores was coming coz of triazine and the non-triazine part. I even mentioned in one of the discussion posts 2 days ago. Everybody was trying their level best to come up with a good model that can generalize well! But, generalizing in biological competitions ain't that easy after all! We see that in real-world as well! :)
Teach me how to learn private test by public LB scores.
", "rawMarkdown": "Teach me how to learn **private** test by **public** LB scores.", "votes": 1, "replies": [ { "id": 2914263, "postDate": "2024-07-09T19:51:24.803Z", "content": "may I say that model perform well in average private test and public lb would be more robust ? given this results, I'm confused.
", "rawMarkdown": "may I say that model perform well in average private test and public lb would be more robust ? given this results, I'm confused." } ] } ] }, { "id": 2912754, "postDate": "2024-07-09T04:07:14.183Z", "content": "This is a virtual win! It happens many times. Its human intuition to select the best model through experience.
\nRefer to my note book on Game theory applications in Machine Learning Competitions:
\nhttps://www.kaggle.com/code/crsuthikshnkumar/gametheory-appln-ml-competitions-crsk
Got the same, luck matters
", "votes": 3, "replies": [ { "id": 2913155, "author_name": "AC", "author_url": "", "post_date": "2024-07-09T09:37:16.850000", "content": "Indeed! After seeing a couple of single submissions in the gold zone, I am sold on the luck part! π
", "votes": 2, "replies": [] } ] }, { "id": 2913217, "author_name": "Matt McDonagh", "author_url": "", "post_date": "2024-07-09T10:35:54.383000", "content": "I enjoyed this Public Score v Private Score dynamic.
\nClearly some folks with 400+ submissions were teaching the test⦠and not generalizing.
", "votes": -2, "replies": [ { "id": 2913228, "author_name": "Antonina Dolgorukova", "author_url": "", "post_date": "2024-07-09T10:46:27", "content": "ahaha, btw, the team that used all the available sibmissions received special congratulations for their submission robustness, showing less variance between public and private scores π
", "votes": 2, "replies": [ { "id": 2913236, "author_name": "AC", "author_url": "", "post_date": "2024-07-09T10:56:17.477000", "content": "Hahaha! Congratulations for another gold! @antoninadolgorukova π₯³
\n@mattmcdonagh I don't think anybody was trying to teach the test thou! :)
\nWe all knew the drop in scores was coming coz of triazine and the non-triazine part. I even mentioned in one of the discussion posts 2 days ago. Everybody was trying their level best to come up with a good model that can generalize well! But, generalizing in biological competitions ain't that easy after all! We see that in real-world as well! :)
Teach me how to learn private test by public LB scores.
", "votes": 1, "replies": [ { "id": 2914263, "author_name": "yuki.fuji", "author_url": "", "post_date": "2024-07-09T19:51:24.803000", "content": "may I say that model perform well in average private test and public lb would be more robust ? given this results, I'm confused.
", "votes": 0, "replies": [] } ] } ] }, { "id": 2912754, "author_name": "C R Suthikshn Kumar", "author_url": "", "post_date": "2024-07-09T04:07:14.183000", "content": "This is a virtual win! It happens many times. Its human intuition to select the best model through experience.
\nRefer to my note book on Game theory applications in Machine Learning Competitions:
\nhttps://www.kaggle.com/code/crsuthikshnkumar/gametheory-appln-ml-competitions-crsk