{ "id": 223704, "title": "Will it have a Big Shake?", "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/223704", "author_name": "", "post_date": "2021-03-05T07:18:45.619740600Z", "votes": 5, "comment_count": 13, "views": 0, "content": "

The Lb is so high and some teams crowd in [970,967]

", "messages": [ { "id": "1227082", "postDate": "03/05/2021 07:18:45", "content": "

The Lb is so high and some teams crowd in [970,967]

", "rawMarkdown": "The Lb is so high and some teams crowd in [970,967]", "votes": null }, { "id": "1227175", "postDate": "03/05/2021 08:57:36", "content": "

yes the shake up will be huge because of low cv and high lb. Most of us are overfitting lb not excluding myself :)

", "rawMarkdown": "yes the shake up will be huge because of low cv and high lb. Most of us are overfitting lb not excluding myself :)", "votes": null }, { "id": "1227190", "postDate": "03/05/2021 09:22:40", "content": "

Clearly, there will be a large shake-up in terms of ranks, because there's little difference in public LB scores for so many, and because of all the public ResNet200D notebooks either going up or down together.

", "rawMarkdown": "Clearly, there will be a large shake-up in terms of ranks, because there's little difference in public LB scores for so many, and because of all the public ResNet200D notebooks either going up or down together.", "votes": null }, { "id": "1227199", "postDate": "03/05/2021 09:42:20", "content": "

I think this competition will be similar to cassava competition, where 4th decimal point + submission time will come to play.
\nWith around 10,000 test images and 4 decimal points means that each image contributes to 0.0001. If you are looking at the current LB, there are more than 200 teams that have score between 0.967-0.969. This means that we have nearly 100 teams with the same 3th decimal points or 10 teams with the same 4th decimal point, so who submit first will have the edge, especially with more than 200+ forked public notebooks at 0.965-0.967.

", "rawMarkdown": "I think this competition will be similar to cassava competition, where 4th decimal point + submission time will come to play.\nWith around 10,000 test images and 4 decimal points means that each image contributes to 0.0001. If you are looking at the current LB, there are more than 200 teams that have score between 0.967-0.969. This means that we have nearly 100 teams with the same 3th decimal points or 10 teams with the same 4th decimal point, so who submit first will have the edge, especially with more than 200+ forked public notebooks at 0.965-0.967.", "votes": null }, { "id": "1227204", "postDate": "03/05/2021 09:50:18", "content": "

they can easily use more digits here

", "rawMarkdown": "they can easily use more digits here", "votes": null }, { "id": "1227205", "postDate": "03/05/2021 09:51:25", "content": "

No, no large shakeup

\n", "rawMarkdown": "No, no large shakeup\n\n- @bestfitting has neven been shaken before\n- i have very high confidence for @Dieter and @philippsinger", "votes": null }, { "id": "1227210", "postDate": "03/05/2021 09:56:07", "content": "

It doesn't make sense to go lower than 4th decimal point because each image contributes to 0.0001 (If total of test images is around 10k). The 5th decimal point will have a minimal impact.

", "rawMarkdown": "It doesn't make sense to go lower than 4th decimal point because each image contributes to 0.0001 (If total of test images is around 10k). The 5th decimal point will have a minimal impact.", "votes": null }, { "id": "1227279", "postDate": "03/05/2021 11:31:45", "content": "

In my opinion, we'll get a shake-up.
\nSo here is my case.
\nGroupKFold:
\n0.9679
\n0.9644
\n0.9665
\n0.9686
\n0.9691
\nAvg: 0.96724
\nLB: 0.961 (TTA), 0.960 (NO TTA)
\nHope to see my real score on private.

", "rawMarkdown": "In my opinion, we'll get a shake-up. \nSo here is my case.\nGroupKFold:\n0.9679\n0.9644\n0.9665\n0.9686\n0.9691\nAvg: 0.96724\nLB: 0.961 (TTA), 0.960 (NO TTA)\nHope to see my real score on private.", "votes": null }, { "id": "1227284", "postDate": "03/05/2021 11:35:07", "content": "

I am also confident in those people who are now in the golden zone, they are unlikely to fall down, except that they will change places. But the remaining participants are not insured in any way.

", "rawMarkdown": "I am also confident in those people who are now in the golden zone, they are unlikely to fall down, except that they will change places. But the remaining participants are not insured in any way.", "votes": null }, { "id": "1227360", "postDate": "03/05/2021 12:56:50", "content": "

If we clear the label by myself , will it benifit the private set?

", "rawMarkdown": "If we clear the label by myself , will it benifit the private set?", "votes": null }, { "id": "1227652", "postDate": "03/05/2021 18:07:17", "content": "

I think there will be a bit of a shuffle within the .967-.969 range, but only small amounts since variation is relatively low.

", "rawMarkdown": "I think there will be a bit of a shuffle within the .967-.969 range, but only small amounts since variation is relatively low.", "votes": null }, { "id": "1238538", "postDate": "03/15/2021 04:50:06", "content": "

The shake not only caused by the auc varience , but also the density of people on lb.

", "rawMarkdown": "The shake not only caused by the auc varience , but also the density of people on lb.", "votes": null }, { "id": "1238820", "postDate": "03/15/2021 10:18:23", "content": "

I was observing a good correlation between CV and LB in the last few weeks, but the relationship became more unstable after reaching 0.969 - 0.970 LB. This makes me think some shakeup in this area is expected.

", "rawMarkdown": "I was observing a good correlation between CV and LB in the last few weeks, but the relationship became more unstable after reaching 0.969 - 0.970 LB. This makes me think some shakeup in this area is expected.", "votes": null }, { "id": "1239652", "postDate": "03/15/2021 22:15:00", "content": "

Large shake up might happen in 96-97 zone. Most of us may just be overfitting public dataset which is just 25% of the total data.

", "rawMarkdown": "Large shake up might happen in 96-97 zone. Most of us may just be overfitting public dataset which is just 25% of the total data.", "votes": null } ], "comments": [ { "id": 1227175, "author_name": "vickygoyal", "author_url": "", "post_date": "03/05/2021 08:57:36", "content": "

yes the shake up will be huge because of low cv and high lb. Most of us are overfitting lb not excluding myself :)

", "votes": null, "replies": [] }, { "id": 1227190, "author_name": "bjoernholzhauer", "author_url": "", "post_date": "03/05/2021 09:22:40", "content": "

Clearly, there will be a large shake-up in terms of ranks, because there's little difference in public LB scores for so many, and because of all the public ResNet200D notebooks either going up or down together.

", "votes": null, "replies": [] }, { "id": 1227199, "author_name": "tom88jerry", "author_url": "", "post_date": "03/05/2021 09:42:20", "content": "

I think this competition will be similar to cassava competition, where 4th decimal point + submission time will come to play.
\nWith around 10,000 test images and 4 decimal points means that each image contributes to 0.0001. If you are looking at the current LB, there are more than 200 teams that have score between 0.967-0.969. This means that we have nearly 100 teams with the same 3th decimal points or 10 teams with the same 4th decimal point, so who submit first will have the edge, especially with more than 200+ forked public notebooks at 0.965-0.967.

", "votes": null, "replies": [ { "id": 1227204, "author_name": "philippsinger", "author_url": "", "post_date": "03/05/2021 09:50:18", "content": "

they can easily use more digits here

", "votes": null, "replies": [] }, { "id": 1227210, "author_name": "tom88jerry", "author_url": "", "post_date": "03/05/2021 09:56:07", "content": "

It doesn't make sense to go lower than 4th decimal point because each image contributes to 0.0001 (If total of test images is around 10k). The 5th decimal point will have a minimal impact.

", "votes": null, "replies": [] }, { "id": 1227360, "author_name": "zekunn", "author_url": "", "post_date": "03/05/2021 12:56:50", "content": "

If we clear the label by myself , will it benifit the private set?

", "votes": null, "replies": [] } ] }, { "id": 1227205, "author_name": "hengck23", "author_url": "", "post_date": "03/05/2021 09:51:25", "content": "

No, no large shakeup

\n", "votes": null, "replies": [ { "id": 1227284, "author_name": "vadimtimakin", "author_url": "", "post_date": "03/05/2021 11:35:07", "content": "

I am also confident in those people who are now in the golden zone, they are unlikely to fall down, except that they will change places. But the remaining participants are not insured in any way.

", "votes": null, "replies": [] }, { "id": 1227652, "author_name": "ryches", "author_url": "", "post_date": "03/05/2021 18:07:17", "content": "

I think there will be a bit of a shuffle within the .967-.969 range, but only small amounts since variation is relatively low.

", "votes": null, "replies": [] }, { "id": 1238538, "author_name": "zekunn", "author_url": "", "post_date": "03/15/2021 04:50:06", "content": "

The shake not only caused by the auc varience , but also the density of people on lb.

", "votes": null, "replies": [] } ] }, { "id": 1227279, "author_name": "vadimtimakin", "author_url": "", "post_date": "03/05/2021 11:31:45", "content": "

In my opinion, we'll get a shake-up.
\nSo here is my case.
\nGroupKFold:
\n0.9679
\n0.9644
\n0.9665
\n0.9686
\n0.9691
\nAvg: 0.96724
\nLB: 0.961 (TTA), 0.960 (NO TTA)
\nHope to see my real score on private.

", "votes": null, "replies": [] }, { "id": 1238820, "author_name": "kozodoi", "author_url": "", "post_date": "03/15/2021 10:18:23", "content": "

I was observing a good correlation between CV and LB in the last few weeks, but the relationship became more unstable after reaching 0.969 - 0.970 LB. This makes me think some shakeup in this area is expected.

", "votes": null, "replies": [] }, { "id": 1239652, "author_name": "vickygoyal", "author_url": "", "post_date": "03/15/2021 22:15:00", "content": "

Large shake up might happen in 96-97 zone. Most of us may just be overfitting public dataset which is just 25% of the total data.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1227082": "The Lb is so high and some teams crowd in [970,967]", "1227175": "yes the shake up will be huge because of low cv and high lb. Most of us are overfitting lb not excluding myself :)", "1227190": "Clearly, there will be a large shake-up in terms of ranks, because there's little difference in public LB scores for so many, and because of all the public ResNet200D notebooks either going up or down together.", "1227199": "I think this competition will be similar to cassava competition, where 4th decimal point + submission time will come to play.\nWith around 10,000 test images and 4 decimal points means that each image contributes to 0.0001. If you are looking at the current LB, there are more than 200 teams that have score between 0.967-0.969. This means that we have nearly 100 teams with the same 3th decimal points or 10 teams with the same 4th decimal point, so who submit first will have the edge, especially with more than 200+ forked public notebooks at 0.965-0.967.", "1227204": "they can easily use more digits here", "1227205": "No, no large shakeup\n\n- @bestfitting has neven been shaken before\n- i have very high confidence for @Dieter and @philippsinger", "1227210": "It doesn't make sense to go lower than 4th decimal point because each image contributes to 0.0001 (If total of test images is around 10k). The 5th decimal point will have a minimal impact.", "1227279": "In my opinion, we'll get a shake-up. \nSo here is my case.\nGroupKFold:\n0.9679\n0.9644\n0.9665\n0.9686\n0.9691\nAvg: 0.96724\nLB: 0.961 (TTA), 0.960 (NO TTA)\nHope to see my real score on private.", "1227284": "I am also confident in those people who are now in the golden zone, they are unlikely to fall down, except that they will change places. But the remaining participants are not insured in any way.", "1227360": "If we clear the label by myself , will it benifit the private set?", "1227652": "I think there will be a bit of a shuffle within the .967-.969 range, but only small amounts since variation is relatively low.", "1238538": "The shake not only caused by the auc varience , but also the density of people on lb.", "1238820": "I was observing a good correlation between CV and LB in the last few weeks, but the relationship became more unstable after reaching 0.969 - 0.970 LB. This makes me think some shakeup in this area is expected.", "1239652": "Large shake up might happen in 96-97 zone. Most of us may just be overfitting public dataset which is just 25% of the total data." }, "source": "meta" }