{ "id": 341920, "title": "9439 in 2 hours(maybe)! Amazing", "url": "/competitions/AI4Code/discussion/341920", "author_name": "", "post_date": "2022-08-04T19:38:49.820208100Z", "votes": 14, "comment_count": 18, "views": 0, "content": "

With only 1 submission, improve from 92+ to 94+. Seems too big gap to catch up. πŸ˜ƒ
\nEven if not just in 2 hours(multiple submissions in short time) still excellent work, and shows we have much room to improve.

", "messages": [ { "id": "1885016", "postDate": "08/04/2022 19:38:49", "content": "

With only 1 submission, improve from 92+ to 94+. Seems too big gap to catch up. πŸ˜ƒ
\nEven if not just in 2 hours(multiple submissions in short time) still excellent work, and shows we have much room to improve.

", "rawMarkdown": "With only 1 submission, improve from 92+ to 94+. Seems too big gap to catch up. πŸ˜ƒ\nEven if not just in 2 hours(multiple submissions in short time) still excellent work, and shows we have much room to improve.", "votes": null }, { "id": "1885069", "postDate": "08/04/2022 21:00:12", "content": "

I love any competitions for such events because you can learn from them and then try approaches in the next competitions. Unfortunately, the final results will be only in 3 months, but I hope some of the participants will share some tricks from their solutions as soon as possible, of course after the final submission deadline, to not destroy LB as few guys…

", "rawMarkdown": "I love any competitions for such events because you can learn from them and then try approaches in the next competitions. Unfortunately, the final results will be only in 3 months, but I hope some of the participants will share some tricks from their solutions as soon as possible, of course after the final submission deadline, to not destroy LB as few guys...", "votes": null }, { "id": "1885398", "postDate": "08/05/2022 06:54:40", "content": "

Game over.πŸ˜‚

", "rawMarkdown": "Game over.πŸ˜‚", "votes": null }, { "id": "1885464", "postDate": "08/05/2022 07:32:18", "content": "

With more ensemble 97+ is possible for him:)

", "rawMarkdown": "With more ensemble 97+ is possible for him:)", "votes": null }, { "id": "1885472", "postDate": "08/05/2022 07:40:39", "content": "

Too many geniuses here, kaggle is so hard.πŸ˜†

", "rawMarkdown": "Too many geniuses here, kaggle is so hard.πŸ˜†", "votes": null }, { "id": "1885581", "postDate": "08/05/2022 09:11:22", "content": "

I love it when someone jumps in front by a large margin. Doesn't happen often, but it is AWESOME! Nice job!

", "rawMarkdown": "I love it when someone jumps in front by a large margin. Doesn't happen often, but it is AWESOME! Nice job!", "votes": null }, { "id": "1886051", "postDate": "08/05/2022 14:33:57", "content": "

I love it when this happens on private LB.

", "rawMarkdown": "I love it when this happens on private LB.", "votes": null }, { "id": "1886075", "postDate": "08/05/2022 14:47:44", "content": "

Think of our situation, only a couple of hundred ranks behind :)
\nmaybe someday….

", "rawMarkdown": "Think of our situation, only a couple of hundred ranks behind :) \nmaybe someday....", "votes": null }, { "id": "1886326", "postDate": "08/05/2022 17:46:00", "content": "

someone already surpassed him. I guess as private LB could contain code which are externally available in non-Kaggle public sources, there could be some leak. But still so high?

\n

Edit:
\n@init27 sorry, i meant public LB, but same argument may be taken forward to private LB also. By Private LB I mean the (pseudo)Private LB which again is nothing but existing Kaggle kernels and there is good chance that some of them may be replicated in Githubs outside. So this leak could extend to private LB also. Of course it wouldnt work for the real final LB which are new set of kernels for next 3 months. I guess those training on external datasets know about this leak and dont mind it, maybe they have strong CVs and their model becomes stronger. However the risk is that the external dataset may not be allowed (grey area).

", "rawMarkdown": "someone already surpassed him. I guess as private LB could contain code which are externally available in non-Kaggle public sources, there could be some leak. But still so high?\n\nEdit: \n@init27 sorry, i meant public LB, but same argument may be taken forward to private LB also. By Private LB I mean the (pseudo)Private LB which again is nothing but existing Kaggle kernels and there is good chance that some of them may be replicated in Githubs outside. So this leak could extend to private LB also. Of course it wouldnt work for the real final LB which are new set of kernels for next 3 months. I guess those training on external datasets know about this leak and dont mind it, maybe they have strong CVs and their model becomes stronger. However the risk is that the external dataset may not be allowed (grey area).", "votes": null }, { "id": "1886372", "postDate": "08/05/2022 18:46:25", "content": "

@allohvk Really interested to know how to improve 0.04+ in one submission.

", "rawMarkdown": "allohvk Really interested to know how to improve 0.04+ in one submission.", "votes": null }, { "id": "1886552", "postDate": "08/06/2022 00:43:25", "content": "

is there any external data overlapped with public LB?
\nor trick?

", "rawMarkdown": "is there any external data overlapped with public LB?\nor trick?", "votes": null }, { "id": "1886561", "postDate": "08/06/2022 01:10:35", "content": "

One more team found the magic it seems πŸ™

", "rawMarkdown": "One more team found the magic it seems πŸ™", "votes": null }, { "id": "1887074", "postDate": "08/06/2022 12:15:11", "content": "

@goldenlock thank you sir

", "rawMarkdown": "goldenlock thank you sir", "votes": null }, { "id": "1887172", "postDate": "08/06/2022 13:31:15", "content": "

I am thinking the same.

", "rawMarkdown": "I am thinking the same.", "votes": null }, { "id": "1887199", "postDate": "08/06/2022 13:59:22", "content": "

This post https://www.kaggle.com/competitions/AI4Code/discussion/338863 mentioning 0.01+ improve with the external kaggle dataset. But I found so much duplicates with contest dataset there.

", "rawMarkdown": "This post [https://www.kaggle.com/competitions/AI4Code/discussion/338863](url) mentioning 0.01+ improve with the external kaggle dataset. But I found so much duplicates with contest dataset there.", "votes": null }, { "id": "1887271", "postDate": "08/06/2022 15:12:21", "content": "
\n

I guess as private LB could contain code which are externally available in non-Kaggle public sources, there could be some leak. But still so high?

\n
\n

@allohvk Did you mean public LB? :)

", "rawMarkdown": "> I guess as private LB could contain code which are externally available in non-Kaggle public sources, there could be some leak. But still so high?\n\n@allohvk Did you mean public LB? :)", "votes": null }, { "id": "1887359", "postDate": "08/06/2022 15:54:45", "content": "

@init27 sorry, i meant public LB, but same argument may be taken forward to private LB also. I corrected above

", "rawMarkdown": "init27 sorry, i meant public LB, but same argument may be taken forward to private LB also. I corrected above", "votes": null }, { "id": "1887738", "postDate": "08/07/2022 01:36:00", "content": "

Thank you very much for editing the original message.

\n

+1! Our team decided not to go ahead with the GitHub datasets for this one due to potential license issues.

\n

Now thinking out loud, maybe we should have investigated it πŸ₯²

", "rawMarkdown": "Thank you very much for editing the original message. \n\n+1! Our team decided not to go ahead with the GitHub datasets for this one due to potential license issues. \n\n\nNow thinking out loud, maybe we should have investigated it πŸ₯²", "votes": null }, { "id": "1887742", "postDate": "08/07/2022 01:41:11", "content": "

I don't use external data too~

", "rawMarkdown": "I don't use external data too~", "votes": null } ], "comments": [ { "id": 1885069, "author_name": "vad13irt", "author_url": "", "post_date": "08/04/2022 21:00:12", "content": "

I love any competitions for such events because you can learn from them and then try approaches in the next competitions. Unfortunately, the final results will be only in 3 months, but I hope some of the participants will share some tricks from their solutions as soon as possible, of course after the final submission deadline, to not destroy LB as few guys…

", "votes": null, "replies": [] }, { "id": 1885398, "author_name": "hydantess", "author_url": "", "post_date": "08/05/2022 06:54:40", "content": "

Game over.πŸ˜‚

", "votes": null, "replies": [ { "id": 1885464, "author_name": "goldenlock", "author_url": "", "post_date": "08/05/2022 07:32:18", "content": "

With more ensemble 97+ is possible for him:)

", "votes": null, "replies": [] }, { "id": 1885472, "author_name": "hydantess", "author_url": "", "post_date": "08/05/2022 07:40:39", "content": "

Too many geniuses here, kaggle is so hard.πŸ˜†

", "votes": null, "replies": [] }, { "id": 1886075, "author_name": "allohvk", "author_url": "", "post_date": "08/05/2022 14:47:44", "content": "

Think of our situation, only a couple of hundred ranks behind :)
\nmaybe someday….

", "votes": null, "replies": [] }, { "id": 1886326, "author_name": "allohvk", "author_url": "", "post_date": "08/05/2022 17:46:00", "content": "

someone already surpassed him. I guess as private LB could contain code which are externally available in non-Kaggle public sources, there could be some leak. But still so high?

\n

Edit:
\n@init27 sorry, i meant public LB, but same argument may be taken forward to private LB also. By Private LB I mean the (pseudo)Private LB which again is nothing but existing Kaggle kernels and there is good chance that some of them may be replicated in Githubs outside. So this leak could extend to private LB also. Of course it wouldnt work for the real final LB which are new set of kernels for next 3 months. I guess those training on external datasets know about this leak and dont mind it, maybe they have strong CVs and their model becomes stronger. However the risk is that the external dataset may not be allowed (grey area).

", "votes": null, "replies": [] }, { "id": 1886372, "author_name": "goldenlock", "author_url": "", "post_date": "08/05/2022 18:46:25", "content": "

@allohvk Really interested to know how to improve 0.04+ in one submission.

", "votes": null, "replies": [] }, { "id": 1886552, "author_name": "hydantess", "author_url": "", "post_date": "08/06/2022 00:43:25", "content": "

is there any external data overlapped with public LB?
\nor trick?

", "votes": null, "replies": [] }, { "id": 1887172, "author_name": "cpmpml", "author_url": "", "post_date": "08/06/2022 13:31:15", "content": "

I am thinking the same.

", "votes": null, "replies": [] }, { "id": 1887199, "author_name": "goldenlock", "author_url": "", "post_date": "08/06/2022 13:59:22", "content": "

This post https://www.kaggle.com/competitions/AI4Code/discussion/338863 mentioning 0.01+ improve with the external kaggle dataset. But I found so much duplicates with contest dataset there.

", "votes": null, "replies": [] }, { "id": 1887271, "author_name": "init27", "author_url": "", "post_date": "08/06/2022 15:12:21", "content": "
\n

I guess as private LB could contain code which are externally available in non-Kaggle public sources, there could be some leak. But still so high?

\n
\n

@allohvk Did you mean public LB? :)

", "votes": null, "replies": [] }, { "id": 1887359, "author_name": "allohvk", "author_url": "", "post_date": "08/06/2022 15:54:45", "content": "

@init27 sorry, i meant public LB, but same argument may be taken forward to private LB also. I corrected above

", "votes": null, "replies": [] }, { "id": 1887738, "author_name": "init27", "author_url": "", "post_date": "08/07/2022 01:36:00", "content": "

Thank you very much for editing the original message.

\n

+1! Our team decided not to go ahead with the GitHub datasets for this one due to potential license issues.

\n

Now thinking out loud, maybe we should have investigated it πŸ₯²

", "votes": null, "replies": [] }, { "id": 1887742, "author_name": "hydantess", "author_url": "", "post_date": "08/07/2022 01:41:11", "content": "

I don't use external data too~

", "votes": null, "replies": [] } ] }, { "id": 1885581, "author_name": "crodoc", "author_url": "", "post_date": "08/05/2022 09:11:22", "content": "

I love it when someone jumps in front by a large margin. Doesn't happen often, but it is AWESOME! Nice job!

", "votes": null, "replies": [ { "id": 1886051, "author_name": "cpmpml", "author_url": "", "post_date": "08/05/2022 14:33:57", "content": "

I love it when this happens on private LB.

", "votes": null, "replies": [] } ] }, { "id": 1886561, "author_name": "init27", "author_url": "", "post_date": "08/06/2022 01:10:35", "content": "

One more team found the magic it seems πŸ™

", "votes": null, "replies": [] }, { "id": 1887074, "author_name": "bilalsuppal", "author_url": "", "post_date": "08/06/2022 12:15:11", "content": "

@goldenlock thank you sir

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1885016": "With only 1 submission, improve from 92+ to 94+. Seems too big gap to catch up. πŸ˜ƒ\nEven if not just in 2 hours(multiple submissions in short time) still excellent work, and shows we have much room to improve.", "1885069": "I love any competitions for such events because you can learn from them and then try approaches in the next competitions. Unfortunately, the final results will be only in 3 months, but I hope some of the participants will share some tricks from their solutions as soon as possible, of course after the final submission deadline, to not destroy LB as few guys...", "1885398": "Game over.πŸ˜‚", "1885464": "With more ensemble 97+ is possible for him:)", "1885472": "Too many geniuses here, kaggle is so hard.πŸ˜†", "1885581": "I love it when someone jumps in front by a large margin. Doesn't happen often, but it is AWESOME! Nice job!", "1886051": "I love it when this happens on private LB.", "1886075": "Think of our situation, only a couple of hundred ranks behind :) \nmaybe someday....", "1886326": "someone already surpassed him. I guess as private LB could contain code which are externally available in non-Kaggle public sources, there could be some leak. But still so high?\n\nEdit: \n@init27 sorry, i meant public LB, but same argument may be taken forward to private LB also. By Private LB I mean the (pseudo)Private LB which again is nothing but existing Kaggle kernels and there is good chance that some of them may be replicated in Githubs outside. So this leak could extend to private LB also. Of course it wouldnt work for the real final LB which are new set of kernels for next 3 months. I guess those training on external datasets know about this leak and dont mind it, maybe they have strong CVs and their model becomes stronger. However the risk is that the external dataset may not be allowed (grey area).", "1886372": "allohvk Really interested to know how to improve 0.04+ in one submission.", "1886552": "is there any external data overlapped with public LB?\nor trick?", "1886561": "One more team found the magic it seems πŸ™", "1887074": "goldenlock thank you sir", "1887172": "I am thinking the same.", "1887199": "This post [https://www.kaggle.com/competitions/AI4Code/discussion/338863](url) mentioning 0.01+ improve with the external kaggle dataset. But I found so much duplicates with contest dataset there.", "1887271": "> I guess as private LB could contain code which are externally available in non-Kaggle public sources, there could be some leak. But still so high?\n\n@allohvk Did you mean public LB? :)", "1887359": "init27 sorry, i meant public LB, but same argument may be taken forward to private LB also. I corrected above", "1887738": "Thank you very much for editing the original message. \n\n+1! Our team decided not to go ahead with the GitHub datasets for this one due to potential license issues. \n\n\nNow thinking out loud, maybe we should have investigated it πŸ₯²", "1887742": "I don't use external data too~" }, "source": "meta" }