{ "id": 84890, "title": "LB saturation", "url": "/competitions/LANL-Earthquake-Prediction/discussion/84890", "author_name": "", "post_date": "2019-03-20T10:26:05.831887100Z", "votes": 7, "comment_count": 9, "views": 0, "content": "

Spots from 83 to 309 have exact same score in the first three decimals. This is taking the concept of LB crowding to a whole new level :-)

", "messages": [ { "id": "494866", "postDate": "03/20/2019 10:26:05", "content": "

Spots from 83 to 309 have exact same score in the first three decimals. This is taking the concept of LB crowding to a whole new level :-)

", "rawMarkdown": "Spots from 83 to 309 have exact same score in the first three decimals. This is taking the concept of LB crowding to a whole new level :-)", "votes": null }, { "id": "495019", "postDate": "03/20/2019 14:43:41", "content": "

That's just the small stuff.\nCheck the Santander leaderboard. That's overcrowding :)

", "rawMarkdown": "That's just the small stuff.\nCheck the Santander leaderboard. That's overcrowding :)", "votes": null }, { "id": "495152", "postDate": "03/20/2019 18:12:00", "content": "

all with the same score as the best public kernel: coincidence?

", "rawMarkdown": "all with the same score as the best public kernel: coincidence?", "votes": null }, { "id": "495381", "postDate": "03/21/2019 02:41:29", "content": "

@konradb, it is now 83 to 320 and counting :-)

", "rawMarkdown": "konradb, it is now 83 to 320 and counting :-)", "votes": null }, { "id": "495476", "postDate": "03/21/2019 06:54:19", "content": "

Wondering how the LB will change when the private test data becomes available :)

", "rawMarkdown": "Wondering how the LB will change when the private test data becomes available :)", "votes": null }, { "id": "496104", "postDate": "03/21/2019 22:36:38", "content": "

I was going to post the same observation but you beat me to it. :)\nIt feels empty here compared to the latest Santander competition. ;)

", "rawMarkdown": "I was going to post the same observation but you beat me to it. :)\nIt feels empty here compared to the latest Santander competition. ;)", "votes": null }, { "id": "497068", "postDate": "03/22/2019 23:04:28", "content": "

I really don't understand why people do that. You're not going to win a medal and more importantly, you don't learn anything either...

", "rawMarkdown": "I really don't understand why people do that. You're not going to win a medal and more importantly, you don't learn anything either...", "votes": null }, { "id": "497320", "postDate": "03/23/2019 11:25:15", "content": "

If the person who downvoted your comment could come and clarify: that would be great :)

", "rawMarkdown": "If the person who downvoted your comment could come and clarify: that would be great :)", "votes": null }, { "id": "499990", "postDate": "03/25/2019 13:36:15", "content": "

Brace yourself, the shake-up is coming

", "rawMarkdown": "Brace yourself, the shake-up is coming", "votes": null }, { "id": "501920", "postDate": "03/28/2019 00:53:09", "content": "

As someone who used it (but has no plans of ever submitting it for the private LB) to test how its features worked as well as verify its LB score, the biggest problem with the (as of right now) best public kernel is that it's a literal black box.

\n\n

We have no idea how the user(s) came away with the coefficients for any of the features, because they were never provided. It's just a giant pre-created function.

\n\n

Had we been able to see how the coefficients were created, it would be a great kernel because it shows that there are other methods out there other than GBTs/RFs/AI/blending/stacking that can compete.

\n\n

As is, the most you can come away with is that there are certain features [all of which follow the exact same set-up as other kernels posted here] that have some significance.

", "rawMarkdown": "As someone who used it (but has no plans of ever submitting it for the private LB) to test how its features worked as well as verify its LB score, the biggest problem with the (as of right now) best public kernel is that it's a literal black box.\n\nWe have no idea how the user(s) came away with the coefficients for any of the features, because they were never provided. It's just a giant pre-created function.\n\nHad we been able to see how the coefficients were created, it would be a great kernel because it shows that there are other methods out there other than GBTs/RFs/AI/blending/stacking that can compete.\n\nAs is, the most you can come away with is that there are certain features [all of which follow the exact same set-up as other kernels posted here] that have some significance.", "votes": null } ], "comments": [ { "id": 495019, "author_name": "stanislavblinov", "author_url": "", "post_date": "03/20/2019 14:43:41", "content": "

That's just the small stuff.\nCheck the Santander leaderboard. That's overcrowding :)

", "votes": null, "replies": [ { "id": 496104, "author_name": "yassinealouini", "author_url": "", "post_date": "03/21/2019 22:36:38", "content": "

I was going to post the same observation but you beat me to it. :)\nIt feels empty here compared to the latest Santander competition. ;)

", "votes": null, "replies": [] } ] }, { "id": 495152, "author_name": "stecasasso", "author_url": "", "post_date": "03/20/2019 18:12:00", "content": "

all with the same score as the best public kernel: coincidence?

", "votes": null, "replies": [ { "id": 497068, "author_name": "bigironsphere", "author_url": "", "post_date": "03/22/2019 23:04:28", "content": "

I really don't understand why people do that. You're not going to win a medal and more importantly, you don't learn anything either...

", "votes": null, "replies": [] }, { "id": 497320, "author_name": "stecasasso", "author_url": "", "post_date": "03/23/2019 11:25:15", "content": "

If the person who downvoted your comment could come and clarify: that would be great :)

", "votes": null, "replies": [] }, { "id": 501920, "author_name": "conormcnamara", "author_url": "", "post_date": "03/28/2019 00:53:09", "content": "

As someone who used it (but has no plans of ever submitting it for the private LB) to test how its features worked as well as verify its LB score, the biggest problem with the (as of right now) best public kernel is that it's a literal black box.

\n\n

We have no idea how the user(s) came away with the coefficients for any of the features, because they were never provided. It's just a giant pre-created function.

\n\n

Had we been able to see how the coefficients were created, it would be a great kernel because it shows that there are other methods out there other than GBTs/RFs/AI/blending/stacking that can compete.

\n\n

As is, the most you can come away with is that there are certain features [all of which follow the exact same set-up as other kernels posted here] that have some significance.

", "votes": null, "replies": [] } ] }, { "id": 495381, "author_name": "sheriytm", "author_url": "", "post_date": "03/21/2019 02:41:29", "content": "

@konradb, it is now 83 to 320 and counting :-)

", "votes": null, "replies": [] }, { "id": 495476, "author_name": "ricarddelgado", "author_url": "", "post_date": "03/21/2019 06:54:19", "content": "

Wondering how the LB will change when the private test data becomes available :)

", "votes": null, "replies": [] }, { "id": 499990, "author_name": "azizbenothman", "author_url": "", "post_date": "03/25/2019 13:36:15", "content": "

Brace yourself, the shake-up is coming

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "494866": "Spots from 83 to 309 have exact same score in the first three decimals. This is taking the concept of LB crowding to a whole new level :-)", "495019": "That's just the small stuff.\nCheck the Santander leaderboard. That's overcrowding :)", "495152": "all with the same score as the best public kernel: coincidence?", "495381": "konradb, it is now 83 to 320 and counting :-)", "495476": "Wondering how the LB will change when the private test data becomes available :)", "496104": "I was going to post the same observation but you beat me to it. :)\nIt feels empty here compared to the latest Santander competition. ;)", "497068": "I really don't understand why people do that. You're not going to win a medal and more importantly, you don't learn anything either...", "497320": "If the person who downvoted your comment could come and clarify: that would be great :)", "499990": "Brace yourself, the shake-up is coming", "501920": "As someone who used it (but has no plans of ever submitting it for the private LB) to test how its features worked as well as verify its LB score, the biggest problem with the (as of right now) best public kernel is that it's a literal black box.\n\nWe have no idea how the user(s) came away with the coefficients for any of the features, because they were never provided. It's just a giant pre-created function.\n\nHad we been able to see how the coefficients were created, it would be a great kernel because it shows that there are other methods out there other than GBTs/RFs/AI/blending/stacking that can compete.\n\nAs is, the most you can come away with is that there are certain features [all of which follow the exact same set-up as other kernels posted here] that have some significance." }, "source": "meta" }