I won't claim I am not disappointed.
\nBut that's it. I selected my best sub, which is also my last one and my best CV one.
\nI guess I was too lucky on public LB early which made me rest too much. Looking forward to reading about all top solutions.
\nCongrats to all those who did well.
", "messages": [ { "id": "1332139", "postDate": "06/02/2021 00:06:32", "content": "I won't claim I am not disappointed.
\nBut that's it. I selected my best sub, which is also my last one and my best CV one.
\nI guess I was too lucky on public LB early which made me rest too much. Looking forward to reading about all top solutions.
\nCongrats to all those who did well.
", "rawMarkdown": "I won't claim I am not disappointed.\n\nBut that's it. I selected my best sub, which is also my last one and my best CV one.\n\nI guess I was too lucky on public LB early which made me rest too much. Looking forward to reading about all top solutions.\n\nCongrats to all those who did well.", "votes": null }, { "id": "1332144", "postDate": "06/02/2021 00:11:25", "content": "A solo gold is always a nice consolation prize :)
\nAny thoughts on why such a shake-up? Did any of your other subs perform better on private?
", "rawMarkdown": "A solo gold is always a nice consolation prize :) \n\nAny thoughts on why such a shake-up? Did any of your other subs perform better on private?", "votes": null }, { "id": "1332146", "postDate": "06/02/2021 00:13:32", "content": "Congratz @cpmpml for the gold! I wish i could share my happiness with you because i got my first medal ever!
\nI also (think) i selected my best model, do we get a few more decimals for our submissions?
\nAlso, how much change can i expect after the hosts verify the results and how long it usually takes?
", "rawMarkdown": "Congratz @cpmpml for the gold! I wish i could share my happiness with you because i got my first medal ever!\n\nI also (think) i selected my best model, do we get a few more decimals for our submissions?\n\nAlso, how much change can i expect after the hosts verify the results and how long it usually takes?", "votes": null }, { "id": "1332154", "postDate": "06/02/2021 00:19:32", "content": "congrats on your first medal! A great feeling indeed. I believe there are many more to come.
\nThe decimals are usually added after a few hours to days.
\nThe final results are also usually published in a matter of days.
Public test set had a noticeably lower optimal threshold than train soundscapes. Different soundscapes also has a huge variation between then, from hundreds of birds to zero.
\nThat was enough for me to not trust both leaderboard and CV and choose two different threshold submissions with a goal to slightly overshoot and slightly undershoot the whole test data, sacrificing a tiny bit of performance if the data would be right between them. Conservative threshold submission actually landed at the exact sweet spot.
\nIt is also interesting to see that real difference between gold zone and my solution was a lot lower than i expected. It was 0.78 to 0.68 in public LB, and it went only to 0.67 to 0.63 on private LB.
\nI started this competiton seriously only 3-4 weeks before the end, so, perhaps, i would had a chance to reach solo gold levels of performance if i had more time. That is inspiring!
\nCongrats @cpmpml with solo gold, and congrats everyone who won medals or got in the money.
", "rawMarkdown": "Public test set had a noticeably lower optimal threshold than train soundscapes. Different soundscapes also has a huge variation between then, from hundreds of birds to zero.\n\nThat was enough for me to not trust both leaderboard and CV and choose two different threshold submissions with a goal to slightly overshoot and slightly undershoot the whole test data, sacrificing a tiny bit of performance if the data would be right between them. Conservative threshold submission actually landed at the exact sweet spot.\n\nIt is also interesting to see that _real_ difference between gold zone and my solution was a lot lower than i expected. It was 0.78 to 0.68 in public LB, and it went only to 0.67 to 0.63 on private LB.\n\nI started this competiton seriously only 3-4 weeks before the end, so, perhaps, i would had a chance to reach solo gold levels of performance if i had more time. That is inspiring!\n\nCongrats @cpmpml with solo gold, and congrats everyone who won medals or got in the money.", "votes": null }, { "id": "1332171", "postDate": "06/02/2021 00:32:00", "content": "Congratulations 🎉 @cpmpml on solo Gold. That's a dream for many of us 😂. Thanks for making the competition interesting with your discussions. You kept everyone on their toes. Took the best out of people. Looking forward to seeing you in future competitions.
", "rawMarkdown": "Congratulations 🎉 @cpmpml on solo Gold. That's a dream for many of us 😂. Thanks for making the competition interesting with your discussions. You kept everyone on their toes. Took the best out of people. Looking forward to seeing you in future competitions.", "votes": null }, { "id": "1332173", "postDate": "06/02/2021 00:34:31", "content": "Congratulations @victorasso on your first medal…
\nAlso congrats @ilu000 for your 2nd place finish… After all 3 weeks was more than enough for your team to get into prize zone :)
I selected my best private score (as most often, if not always) But I ovdrfit still. I think I know why. Too late
", "rawMarkdown": "I selected my best private score (as most often, if not always) But I ovdrfit still. I think I know why. Too late", "votes": null }, { "id": "1332177", "postDate": "06/02/2021 00:40:33", "content": "Congrats @cpmpml with solo gold, and congrats everyone who won medals.
\nI think it’s a pleasure to choose the best private LB model.
\nEnjoy the competition and keep learning
Thank you! Congratz to you too for the gold!
\nMay i ask, what do you consider it was the best idea you guys had to get you up there?
", "rawMarkdown": "Thank you! Congratz to you too for the gold! \n\nMay i ask, what do you consider it was the best idea you guys had to get you up there?", "votes": null }, { "id": "1332376", "postDate": "06/02/2021 05:18:01", "content": "Congratulations on the solo gold, @cpmpml!! The best one can do is trust your CV and judgement… there will be ups and downs.. and the private LB has will have some shakeups, some unfavourable.. hopefully many more that are favourable.. as you have also said many times, best is to trust the CV and that is one wonderful learning from Kaggle..
\na few minor shakeups are a distraction and noise in the process
thanks for all the sharing across the competition
", "rawMarkdown": "Congratulations on the solo gold, @cpmpml!! The best one can do is trust your CV and judgement... there will be ups and downs.. and the private LB has will have some shakeups, some unfavourable.. hopefully many more that are favourable.. as you have also said many times, best is to trust the CV and that is one wonderful learning from Kaggle..\na few minor shakeups are a distraction and noise in the process\n\nthanks for all the sharing across the competition", "votes": null }, { "id": "1332693", "postDate": "06/02/2021 08:44:22", "content": "Not to forget, your contribution in the competition is more than a score, comment and helping with insight in almost every relevant topic, honorable, thank you!
", "rawMarkdown": "Not to forget, your contribution in the competition is more than a score, comment and helping with insight in almost every relevant topic, honorable, thank you!", "votes": null }, { "id": "1333564", "postDate": "06/02/2021 21:17:41", "content": "Congrats @cpmpml with solo gold,👍
", "rawMarkdown": "Congrats @cpmpml with solo gold,👍", "votes": null } ], "comments": [ { "id": 1332144, "author_name": "leecming", "author_url": "", "post_date": "06/02/2021 00:11:25", "content": "A solo gold is always a nice consolation prize :)
\nAny thoughts on why such a shake-up? Did any of your other subs perform better on private?
", "votes": null, "replies": [ { "id": 1332175, "author_name": "cpmpml", "author_url": "", "post_date": "06/02/2021 00:37:30", "content": "I selected my best private score (as most often, if not always) But I ovdrfit still. I think I know why. Too late
", "votes": null, "replies": [] } ] }, { "id": 1332146, "author_name": "victorasso", "author_url": "", "post_date": "06/02/2021 00:13:32", "content": "Congratz @cpmpml for the gold! I wish i could share my happiness with you because i got my first medal ever!
\nI also (think) i selected my best model, do we get a few more decimals for our submissions?
\nAlso, how much change can i expect after the hosts verify the results and how long it usually takes?
", "votes": null, "replies": [ { "id": 1332154, "author_name": "ilu000", "author_url": "", "post_date": "06/02/2021 00:19:32", "content": "congrats on your first medal! A great feeling indeed. I believe there are many more to come.
\nThe decimals are usually added after a few hours to days.
\nThe final results are also usually published in a matter of days.
Congratulations @victorasso on your first medal…
\nAlso congrats @ilu000 for your 2nd place finish… After all 3 weeks was more than enough for your team to get into prize zone :)
Thank you! Congratz to you too for the gold!
\nMay i ask, what do you consider it was the best idea you guys had to get you up there?
", "votes": null, "replies": [] } ] }, { "id": 1332160, "author_name": "fffrrt", "author_url": "", "post_date": "06/02/2021 00:26:02", "content": "Public test set had a noticeably lower optimal threshold than train soundscapes. Different soundscapes also has a huge variation between then, from hundreds of birds to zero.
\nThat was enough for me to not trust both leaderboard and CV and choose two different threshold submissions with a goal to slightly overshoot and slightly undershoot the whole test data, sacrificing a tiny bit of performance if the data would be right between them. Conservative threshold submission actually landed at the exact sweet spot.
\nIt is also interesting to see that real difference between gold zone and my solution was a lot lower than i expected. It was 0.78 to 0.68 in public LB, and it went only to 0.67 to 0.63 on private LB.
\nI started this competiton seriously only 3-4 weeks before the end, so, perhaps, i would had a chance to reach solo gold levels of performance if i had more time. That is inspiring!
\nCongrats @cpmpml with solo gold, and congrats everyone who won medals or got in the money.
", "votes": null, "replies": [] }, { "id": 1332171, "author_name": "lplenka", "author_url": "", "post_date": "06/02/2021 00:32:00", "content": "Congratulations 🎉 @cpmpml on solo Gold. That's a dream for many of us 😂. Thanks for making the competition interesting with your discussions. You kept everyone on their toes. Took the best out of people. Looking forward to seeing you in future competitions.
", "votes": null, "replies": [] }, { "id": 1332177, "author_name": "whutddmm", "author_url": "", "post_date": "06/02/2021 00:40:33", "content": "Congrats @cpmpml with solo gold, and congrats everyone who won medals.
\nI think it’s a pleasure to choose the best private LB model.
\nEnjoy the competition and keep learning
Congratulations on the solo gold, @cpmpml!! The best one can do is trust your CV and judgement… there will be ups and downs.. and the private LB has will have some shakeups, some unfavourable.. hopefully many more that are favourable.. as you have also said many times, best is to trust the CV and that is one wonderful learning from Kaggle..
\na few minor shakeups are a distraction and noise in the process
thanks for all the sharing across the competition
", "votes": null, "replies": [] }, { "id": 1332693, "author_name": "kirderf", "author_url": "", "post_date": "06/02/2021 08:44:22", "content": "Not to forget, your contribution in the competition is more than a score, comment and helping with insight in almost every relevant topic, honorable, thank you!
", "votes": null, "replies": [] }, { "id": 1333564, "author_name": "mohamedbakrey", "author_url": "", "post_date": "06/02/2021 21:17:41", "content": "Congrats @cpmpml with solo gold,👍
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1332139": "I won't claim I am not disappointed.\n\nBut that's it. I selected my best sub, which is also my last one and my best CV one.\n\nI guess I was too lucky on public LB early which made me rest too much. Looking forward to reading about all top solutions.\n\nCongrats to all those who did well.", "1332144": "A solo gold is always a nice consolation prize :) \n\nAny thoughts on why such a shake-up? Did any of your other subs perform better on private?", "1332146": "Congratz @cpmpml for the gold! I wish i could share my happiness with you because i got my first medal ever!\n\nI also (think) i selected my best model, do we get a few more decimals for our submissions?\n\nAlso, how much change can i expect after the hosts verify the results and how long it usually takes?", "1332154": "congrats on your first medal! A great feeling indeed. I believe there are many more to come.\n\nThe decimals are usually added after a few hours to days. \nThe final results are also usually published in a matter of days.", "1332160": "Public test set had a noticeably lower optimal threshold than train soundscapes. Different soundscapes also has a huge variation between then, from hundreds of birds to zero.\n\nThat was enough for me to not trust both leaderboard and CV and choose two different threshold submissions with a goal to slightly overshoot and slightly undershoot the whole test data, sacrificing a tiny bit of performance if the data would be right between them. Conservative threshold submission actually landed at the exact sweet spot.\n\nIt is also interesting to see that _real_ difference between gold zone and my solution was a lot lower than i expected. It was 0.78 to 0.68 in public LB, and it went only to 0.67 to 0.63 on private LB.\n\nI started this competiton seriously only 3-4 weeks before the end, so, perhaps, i would had a chance to reach solo gold levels of performance if i had more time. That is inspiring!\n\nCongrats @cpmpml with solo gold, and congrats everyone who won medals or got in the money.", "1332171": "Congratulations 🎉 @cpmpml on solo Gold. That's a dream for many of us 😂. Thanks for making the competition interesting with your discussions. You kept everyone on their toes. Took the best out of people. Looking forward to seeing you in future competitions.", "1332173": "Congratulations @victorasso on your first medal... \nAlso congrats @ilu000 for your 2nd place finish... After all 3 weeks was more than enough for your team to get into prize zone :)", "1332175": "I selected my best private score (as most often, if not always) But I ovdrfit still. I think I know why. Too late", "1332177": "Congrats @cpmpml with solo gold, and congrats everyone who won medals.\nI think it’s a pleasure to choose the best private LB model.\nEnjoy the competition and keep learning", "1332179": "Thank you! Congratz to you too for the gold! \n\nMay i ask, what do you consider it was the best idea you guys had to get you up there?", "1332376": "Congratulations on the solo gold, @cpmpml!! The best one can do is trust your CV and judgement... there will be ups and downs.. and the private LB has will have some shakeups, some unfavourable.. hopefully many more that are favourable.. as you have also said many times, best is to trust the CV and that is one wonderful learning from Kaggle..\na few minor shakeups are a distraction and noise in the process\n\nthanks for all the sharing across the competition", "1332693": "Not to forget, your contribution in the competition is more than a score, comment and helping with insight in almost every relevant topic, honorable, thank you!", "1333564": "Congrats @cpmpml with solo gold,👍" }, "source": "meta" }