{ "id": 16342, "title": "0.99811? Is that possible?", "url": "/competitions/dato-native/discussion/16342", "author_name": "", "post_date": "2015-09-06T13:45:50.880Z", "votes": 5, "comment_count": 16, "views": 2657, "content": "

I don't believe this is for real, lol.

", "messages": [ { "id": "91696", "postDate": "09/06/2015 13:45:50", "content": "

I don't believe this is for real, lol.

", "rawMarkdown": "I don't believe this is for real, lol.", "votes": null }, { "id": "91702", "postDate": "09/06/2015 15:09:44", "content": "

Lol

", "rawMarkdown": "Lol", "votes": null }, { "id": "91710", "postDate": "09/06/2015 16:20:57", "content": "

[quote=Abhishek;91702]\nLol\n[/quote]

\n\n

Exactly my first thought (lie...) :D

\n\n

Anyway, I think anything over 0.90 AUC, its just for kaggle-only, real life will really get all our scores down.

\n\n

The game has changed now XD

", "rawMarkdown": "[quote=Abhishek;91702]\r\nLol\r\n[/quote]\r\n\r\nExactly my first thought (lie...) :D\r\n\r\nAnyway, I think anything over 0.90 AUC, its just for kaggle-only, real life will really get all our scores down.\r\n\r\nThe game has changed now XD", "votes": null }, { "id": "91717", "postDate": "09/06/2015 16:38:50", "content": "

Whoever thinks this competition will end with a perfect 1.00000, vote up this comment. (just kidding, you don't really need to :P)

", "rawMarkdown": "Whoever thinks this competition will end with a perfect 1.00000, vote up this comment. (just kidding, you don't really need to :P)", "votes": null }, { "id": "91730", "postDate": "09/06/2015 20:55:44", "content": "

is 0.99811 probabilistic ? :-) there exists a deterministic rule to determine Sponsored pgs !!

", "rawMarkdown": "is 0.99811 probabilistic ? :-) there exists a deterministic rule to determine Sponsored pgs !!", "votes": null }, { "id": "91770", "postDate": "09/07/2015 12:24:49", "content": "

I dont think there is any reason now to take part. Its impossible to reach score higher than 0.99811 :P :P

", "rawMarkdown": "I dont think there is any reason now to take part. Its impossible to reach score higher than 0.99811 :P :P", "votes": null }, { "id": "91779", "postDate": "09/07/2015 15:43:22", "content": "

Maybe that's not possible to reach 1 because of errors in some files, but there would be instead a perfect predictor tag, don't you think?

", "rawMarkdown": "Maybe that's not possible to reach 1 because of errors in some files, but there would be instead a perfect predictor tag, don't you think?", "votes": null }, { "id": "91780", "postDate": "09/07/2015 15:56:18", "content": "

At what point does this become target leakage that triggers, hopefully, a restart? Or is this a larger scale rule induction competition?

", "rawMarkdown": "At what point does this become target leakage that triggers, hopefully, a restart? Or is this a larger scale rule induction competition?", "votes": null }, { "id": "91782", "postDate": "09/07/2015 16:18:13", "content": "

It's not a leakage if it is naturally present in the dataset from the start. It was just something hidden to our eyes from the beginning. That's the very purpose of ML that showing those kind of relationships, actually.

\n\n

Why can't people, even in the field, look to this without being suspicious?\nI am not participating to the competition but was interested in some of its aspects, and for working regularly with all kind of html pages and parsing these, I can say that there are a whole lot of things that can tell you stories.

\n\n

Especially when a lot of people don't really code and use templates and tools that build generic code, with the whole lot of formatting to follow good practices and so on.

\n\n

Maybe it's some kind of frame, maybe it is a link or a code for robots or so on?

", "rawMarkdown": "It's not a leakage if it is naturally present in the dataset from the start. It was just something hidden to our eyes from the beginning. That's the very purpose of ML that showing those kind of relationships, actually.\r\n\r\nWhy can't people, even in the field, look to this without being suspicious?\r\nI am not participating to the competition but was interested in some of its aspects, and for working regularly with all kind of html pages and parsing these, I can say that there are a whole lot of things that can tell you stories. \r\n\r\nEspecially when a lot of people don't really code and use templates and tools that build generic code, with the whole lot of formatting to follow good practices and so on.\r\n\r\nMaybe it's some kind of frame, maybe it is a link or a code for robots or so on?", "votes": null }, { "id": "91783", "postDate": "09/07/2015 17:37:02", "content": "

why not possible? I see not much difference between 0.97 and 0.99 :P

\n\n

By the way, a 1 auc is easier than 0 log loss, right ?

", "rawMarkdown": "why not possible? I see not much difference between 0.97 and 0.99 :P\r\n\r\nBy the way, a 1 auc is easier than 0 log loss, right ?", "votes": null }, { "id": "91785", "postDate": "09/07/2015 17:42:00", "content": "

1auc is equivalent to 0 log loss ! just a matter of determining the threshold.

", "rawMarkdown": "1auc is equivalent to 0 log loss ! just a matter of determining the threshold.", "votes": null }, { "id": "91808", "postDate": "09/07/2015 23:07:31", "content": "

There can be many ways to achieve 1 auc but there is one way to get 0 logloss.

", "rawMarkdown": "There can be many ways to achieve 1 auc but there is one way to get 0 logloss.", "votes": null }, { "id": "91833", "postDate": "09/08/2015 08:52:16", "content": "

Actually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?

", "rawMarkdown": "Actually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?", "votes": null }, { "id": "91838", "postDate": "09/08/2015 10:43:32", "content": "

can't you predict it by other approaches?

", "rawMarkdown": "can't you predict it by other approaches?", "votes": null }, { "id": "91854", "postDate": "09/08/2015 14:47:11", "content": "

[quote=Gerard Toonstra;91833]

\n\n

Actually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?

\n\n

[/quote]\nCrowdsource it?

\n\n

Offer some sort of reward to whoever can predict it for you.

", "rawMarkdown": "[quote=Gerard Toonstra;91833]\r\n\r\nActually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?\r\n\r\n[/quote]\r\nCrowdsource it? \r\n\r\nOffer some sort of reward to whoever can predict it for you.", "votes": null }, { "id": "91856", "postDate": "09/08/2015 15:06:27", "content": "

[quote=Gerard Toonstra;91833]

\n\n

Actually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?

\n\n

[/quote]

\n\n

It sounds like a leak. So please either make it public or keep it to yourself : )

", "rawMarkdown": "[quote=Gerard Toonstra;91833]\r\n\r\nActually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?\r\n\r\n[/quote]\r\n\r\nIt sounds like a leak. So please either make it public or keep it to yourself : )", "votes": null }, { "id": "91859", "postDate": "09/08/2015 15:20:11", "content": "

Great responses, but I wasn't being serious. I was referring to the "sponsored" column in an absurd kind of way.

\n\n

Crowd sourcing sounds great though. I should be able to start a competition on kaggle. I offer $4000 in rewards. Competition ends 1st october.

", "rawMarkdown": "Great responses, but I wasn't being serious. I was referring to the \"sponsored\" column in an absurd kind of way.\r\n\r\nCrowd sourcing sounds great though. I should be able to start a competition on kaggle. I offer $4000 in rewards. Competition ends 1st october.", "votes": null } ], "comments": [ { "id": 91702, "author_name": "abhishek", "author_url": "", "post_date": "09/06/2015 15:09:44", "content": "

Lol

", "votes": null, "replies": [] }, { "id": 91710, "author_name": "carloshuertas", "author_url": "", "post_date": "09/06/2015 16:20:57", "content": "

[quote=Abhishek;91702]\nLol\n[/quote]

\n\n

Exactly my first thought (lie...) :D

\n\n

Anyway, I think anything over 0.90 AUC, its just for kaggle-only, real life will really get all our scores down.

\n\n

The game has changed now XD

", "votes": null, "replies": [] }, { "id": 91717, "author_name": "mandalsubhajit", "author_url": "", "post_date": "09/06/2015 16:38:50", "content": "

Whoever thinks this competition will end with a perfect 1.00000, vote up this comment. (just kidding, you don't really need to :P)

", "votes": null, "replies": [] }, { "id": 91730, "author_name": "zsaurk", "author_url": "", "post_date": "09/06/2015 20:55:44", "content": "

is 0.99811 probabilistic ? :-) there exists a deterministic rule to determine Sponsored pgs !!

", "votes": null, "replies": [] }, { "id": 91770, "author_name": "anujmehra", "author_url": "", "post_date": "09/07/2015 12:24:49", "content": "

I dont think there is any reason now to take part. Its impossible to reach score higher than 0.99811 :P :P

", "votes": null, "replies": [] }, { "id": 91779, "author_name": "wildwizard", "author_url": "", "post_date": "09/07/2015 15:43:22", "content": "

Maybe that's not possible to reach 1 because of errors in some files, but there would be instead a perfect predictor tag, don't you think?

", "votes": null, "replies": [] }, { "id": 91780, "author_name": "adjgiulio", "author_url": "", "post_date": "09/07/2015 15:56:18", "content": "

At what point does this become target leakage that triggers, hopefully, a restart? Or is this a larger scale rule induction competition?

", "votes": null, "replies": [] }, { "id": 91782, "author_name": "wildwizard", "author_url": "", "post_date": "09/07/2015 16:18:13", "content": "

It's not a leakage if it is naturally present in the dataset from the start. It was just something hidden to our eyes from the beginning. That's the very purpose of ML that showing those kind of relationships, actually.

\n\n

Why can't people, even in the field, look to this without being suspicious?\nI am not participating to the competition but was interested in some of its aspects, and for working regularly with all kind of html pages and parsing these, I can say that there are a whole lot of things that can tell you stories.

\n\n

Especially when a lot of people don't really code and use templates and tools that build generic code, with the whole lot of formatting to follow good practices and so on.

\n\n

Maybe it's some kind of frame, maybe it is a link or a code for robots or so on?

", "votes": null, "replies": [] }, { "id": 91783, "author_name": "jiweiliu", "author_url": "", "post_date": "09/07/2015 17:37:02", "content": "

why not possible? I see not much difference between 0.97 and 0.99 :P

\n\n

By the way, a 1 auc is easier than 0 log loss, right ?

", "votes": null, "replies": [] }, { "id": 91785, "author_name": "zsaurk", "author_url": "", "post_date": "09/07/2015 17:42:00", "content": "

1auc is equivalent to 0 log loss ! just a matter of determining the threshold.

", "votes": null, "replies": [] }, { "id": 91808, "author_name": "senbong", "author_url": "", "post_date": "09/07/2015 23:07:31", "content": "

There can be many ways to achieve 1 auc but there is one way to get 0 logloss.

", "votes": null, "replies": [] }, { "id": 91833, "author_name": "remap1", "author_url": "", "post_date": "09/08/2015 08:52:16", "content": "

Actually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?

", "votes": null, "replies": [] }, { "id": 91838, "author_name": "wildwizard", "author_url": "", "post_date": "09/08/2015 10:43:32", "content": "

can't you predict it by other approaches?

", "votes": null, "replies": [] }, { "id": 91854, "author_name": "hillbert", "author_url": "", "post_date": "09/08/2015 14:47:11", "content": "

[quote=Gerard Toonstra;91833]

\n\n

Actually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?

\n\n

[/quote]\nCrowdsource it?

\n\n

Offer some sort of reward to whoever can predict it for you.

", "votes": null, "replies": [] }, { "id": 91856, "author_name": "jiweiliu", "author_url": "", "post_date": "09/08/2015 15:06:27", "content": "

[quote=Gerard Toonstra;91833]

\n\n

Actually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?

\n\n

[/quote]

\n\n

It sounds like a leak. So please either make it public or keep it to yourself : )

", "votes": null, "replies": [] }, { "id": 91859, "author_name": "remap1", "author_url": "", "post_date": "09/08/2015 15:20:11", "content": "

Great responses, but I wasn't being serious. I was referring to the "sponsored" column in an absurd kind of way.

\n\n

Crowd sourcing sounds great though. I should be able to start a competition on kaggle. I offer $4000 in rewards. Competition ends 1st october.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "91696": "I don't believe this is for real, lol.", "91702": "Lol", "91710": "[quote=Abhishek;91702]\r\nLol\r\n[/quote]\r\n\r\nExactly my first thought (lie...) :D\r\n\r\nAnyway, I think anything over 0.90 AUC, its just for kaggle-only, real life will really get all our scores down.\r\n\r\nThe game has changed now XD", "91717": "Whoever thinks this competition will end with a perfect 1.00000, vote up this comment. (just kidding, you don't really need to :P)", "91730": "is 0.99811 probabilistic ? :-) there exists a deterministic rule to determine Sponsored pgs !!", "91770": "I dont think there is any reason now to take part. Its impossible to reach score higher than 0.99811 :P :P", "91779": "Maybe that's not possible to reach 1 because of errors in some files, but there would be instead a perfect predictor tag, don't you think?", "91780": "At what point does this become target leakage that triggers, hopefully, a restart? Or is this a larger scale rule induction competition?", "91782": "It's not a leakage if it is naturally present in the dataset from the start. It was just something hidden to our eyes from the beginning. That's the very purpose of ML that showing those kind of relationships, actually.\r\n\r\nWhy can't people, even in the field, look to this without being suspicious?\r\nI am not participating to the competition but was interested in some of its aspects, and for working regularly with all kind of html pages and parsing these, I can say that there are a whole lot of things that can tell you stories. \r\n\r\nEspecially when a lot of people don't really code and use templates and tools that build generic code, with the whole lot of formatting to follow good practices and so on.\r\n\r\nMaybe it's some kind of frame, maybe it is a link or a code for robots or so on?", "91783": "why not possible? I see not much difference between 0.97 and 0.99 :P\r\n\r\nBy the way, a 1 auc is easier than 0 log loss, right ?", "91785": "1auc is equivalent to 0 log loss ! just a matter of determining the threshold.", "91808": "There can be many ways to achieve 1 auc but there is one way to get 0 logloss.", "91833": "Actually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?", "91838": "can't you predict it by other approaches?", "91854": "[quote=Gerard Toonstra;91833]\r\n\r\nActually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?\r\n\r\n[/quote]\r\nCrowdsource it? \r\n\r\nOffer some sort of reward to whoever can predict it for you.", "91856": "[quote=Gerard Toonstra;91833]\r\n\r\nActually, there's a single column in the training data that gives me a perfect score. The column doesn't seem to be included in the test set though. Does anyone know where I can download the column for the test set? I tried to recover the column using some of the scripts found on this site, but it never gives me a perfect score. Please help?\r\n\r\n[/quote]\r\n\r\nIt sounds like a leak. So please either make it public or keep it to yourself : )", "91859": "Great responses, but I wasn't being serious. I was referring to the \"sponsored\" column in an absurd kind of way.\r\n\r\nCrowd sourcing sounds great though. I should be able to start a competition on kaggle. I offer $4000 in rewards. Competition ends 1st october." }, "source": "meta" }