{ "id": 2908, "title": "Unofficial Leaderboard", "url": "/competitions/MerckActivity/discussion/2908", "author_name": "", "post_date": "2012-10-17T03:02:01.297Z", "votes": null, "comment_count": 15, "views": 8490, "content": "

Any one else curious enough to create an unofficial private leaderboard?  

", "messages": [ { "id": "15768", "postDate": "10/17/2012 03:02:01", "content": "

Any one else curious enough to create an unofficial private leaderboard?  

", "rawMarkdown": "", "votes": null }, { "id": "15773", "postDate": "10/17/2012 03:59:45", "content": "

Neil my Friend,

\r\n

Curious but how do we do that?

", "rawMarkdown": "", "votes": null }, { "id": "15779", "postDate": "10/17/2012 04:51:53", "content": "If the top players post their best private scores, we can make an unofficial private leaderboard :)", "rawMarkdown": "", "votes": null }, { "id": "15780", "postDate": "10/17/2012 05:00:07", "content": "

My public score was 0.46984, private score was 0.47018 and my best private model which I didn't submit was 0.47032. I'm was so excited to know the final standings today, but I guess we'll have to wait.

\r\n

Overall, it was great fun for my first competition and I'm pretty happy that I got over the 0.47 barrier.\r\n
\r\nLooking forward to competing again!

", "rawMarkdown": "", "votes": null }, { "id": "15782", "postDate": "10/17/2012 05:26:45", "content": "

Neil - excellent idea. The top prv score for our team was 0.48154.

", "rawMarkdown": "", "votes": null }, { "id": "15784", "postDate": "10/17/2012 05:35:07", "content": "

Yes,because we all have the private score now,if everyone post their private score we can make an unofficial leaderboard. my best public score is 0.47288(public)->0.46907(private),and best private score is :0.46686(public)->0.46931(private).there is a huge\r\n gap between this two model in public score,but the 0.46686 model have a little bit higher than 0.47288 model in private score.BTW,I'm also curious what is your best single model.The above two model are both single model,not using any ensemble methods.

", "rawMarkdown": "", "votes": null }, { "id": "15788", "postDate": "10/17/2012 05:54:11", "content": "

my best submitted private score is 0.47704.

", "rawMarkdown": "", "votes": null }, { "id": "15792", "postDate": "10/17/2012 07:56:26", "content": "

I am not in this competition but I noticed that users kaggle rating points have already been modified according to their private scores in this competition, so if you knew what your kaggle score was before and after the competition finished you could reverse\r\n out your position. From the posts on the forum at the time when the score function was being debated it emerged the points are proportional to log10(competitors)/team_size*position^-0.75. I recently won a competition with 80 players in and got around 200 pts.\r\n From that we can see the winner of this competition will have received around 250 pts, 2nd 148pts, 3rd 109 and 4th around 88. So for example i believe Gxav score before this comp ended was around 555 and it is now around 590. That is he has picked up 35 pts\r\n from a team of 3, which scales up to 105 pts, indicating most likely that team DataRobot finished in 3rd place on the private leaderboard. Berry has a score of 155 so he can not have won the private, he must have finished second. Ruslan is part of team gggg\r\n and has a score of 45. He has only entered one competition and that team has 5 members in, so the team score is around 225pt. Ergo Team gggg won.

\r\n

tl;dr I predict private leaderboard is

\r\n

1. gggg

\r\n

2. Berry

\r\n

3. DataRobot

", "rawMarkdown": "", "votes": null }, { "id": "15802", "postDate": "10/17/2012 10:13:21", "content": "

[quote=Jason Tigg;15792]

\r\n

I am not in this competition but I noticed that users kaggle rating points have already been modified according to their private scores in this competition, 

\r\n

[/quote]

\r\n

Wouldn't that just be the public score? In my user profiled, it had assumed the public as the final standing, hence scores calculated from there? no? 

\r\n

In either case, I predict: 

\r\n

1) Data Robot 

\r\n

2) Berry

", "rawMarkdown": "", "votes": null }, { "id": "15804", "postDate": "10/17/2012 10:24:02", "content": "

Its not consistent with public scores as I explained.

", "rawMarkdown": "", "votes": null }, { "id": "15806", "postDate": "10/17/2012 11:50:46", "content": "

[quote=Jason Tigg;15792]

\r\n

I am not in this competition but I noticed that users kaggle rating points have already been modified according to their private scores in this competition, so if you knew what your kaggle score was before and after the competition finished you could reverse\r\n out your position. From the posts on the forum at the time when the score function was being debated it emerged the points are proportional to log10(competitors)/team_size*position^-0.75. I recently won a competition with 80 players in and got around 200 pts.\r\n From that we can see the winner of this competition will have received around 250 pts, 2nd 148pts, 3rd 109 and 4th around 88. So for example i believe Gxav score before this comp ended was around 555 and it is now around 590. That is he has picked up 35 pts\r\n from a team of 3, which scales up to 105 pts, indicating most likely that team DataRobot finished in 3rd place on the private leaderboard. Berry has a score of 155 so he can not have won the private, he must have finished second. Ruslan is part of team gggg\r\n and has a score of 45. He has only entered one competition and that team has 5 members in, so the team score is around 225pt. Ergo Team gggg won.

\r\n

tl;dr I predict private leaderboard is

\r\n

1. gggg

\r\n

2. Berry

\r\n

3. DataRobot

\r\n

[/quote]

\r\n

Looks logical, then Grangnam Style is #5 and Luxtorpedia is #6.

\r\n

Small discrepancy is in Berry's total score. He has little bit less points than I would expect taking into account his another competition
\r\n(even with time decay).

\r\n

I cannot reject order:

\r\n

1. gggg

\r\n

2. DataRobot

\r\n

3. Berry

\r\n

 4.

\r\n

5. Grangnam Style

\r\n

6.  Luxtorpedia

\r\n

 

\r\n

 

", "rawMarkdown": "", "votes": null }, { "id": "15807", "postDate": "10/17/2012 11:57:35", "content": "

I believe given their ephemeral nature that hackathons are weighted like 1/3 of normal competitions, which would explain what I will refer to as the "Berry Anomaly"

", "rawMarkdown": "", "votes": null }, { "id": "15808", "postDate": "10/17/2012 12:00:52", "content": "

[quote=Jason Tigg;15807]

\r\n

I believe given their ephemeral nature that hackathons are weighted like 1/3 of normal competitions, which would explain what I will refer to as the "Berry Anomaly"

\r\n

[/quote]

\r\n

OK, I probably missed information about that 1/3 coefficient

\r\n

P.S. I found it, per Ben's message in "Users ranking methods?"   hackathrons are weighted as %25

", "rawMarkdown": "", "votes": null }, { "id": "15816", "postDate": "10/17/2012 14:42:53", "content": "

sounds logical - given that different folks have different points. I was 144 before this and have moved to 186 which means I got 42 points.
\r\nHowever this is consistent with the public leaderboard rank of 7

", "rawMarkdown": "", "votes": null }, { "id": "15817", "postDate": "10/17/2012 14:43:54", "content": "

don't understand why private leaderboard results cannot be published - what is it that is being verified now that could not have been done earlier?

", "rawMarkdown": "", "votes": null }, { "id": "15863", "postDate": "10/18/2012 00:33:42", "content": "

"Jason Tigg" data leak was closed. Ranking is back to precompetition level.

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 15773, "author_name": "rkirana", "author_url": "", "post_date": "10/17/2012 03:59:45", "content": "

Neil my Friend,

\r\n

Curious but how do we do that?

", "votes": null, "replies": [] }, { "id": 15779, "author_name": "srgodbole", "author_url": "", "post_date": "10/17/2012 04:51:53", "content": "If the top players post their best private scores, we can make an unofficial private leaderboard :)", "votes": null, "replies": [] }, { "id": 15780, "author_name": "dmitrim", "author_url": "", "post_date": "10/17/2012 05:00:07", "content": "

My public score was 0.46984, private score was 0.47018 and my best private model which I didn't submit was 0.47032. I'm was so excited to know the final standings today, but I guess we'll have to wait.

\r\n

Overall, it was great fun for my first competition and I'm pretty happy that I got over the 0.47 barrier.\r\n
\r\nLooking forward to competing again!

", "votes": null, "replies": [] }, { "id": 15782, "author_name": "konradb", "author_url": "", "post_date": "10/17/2012 05:26:45", "content": "

Neil - excellent idea. The top prv score for our team was 0.48154.

", "votes": null, "replies": [] }, { "id": 15784, "author_name": "stupidbird", "author_url": "", "post_date": "10/17/2012 05:35:07", "content": "

Yes,because we all have the private score now,if everyone post their private score we can make an unofficial leaderboard. my best public score is 0.47288(public)->0.46907(private),and best private score is :0.46686(public)->0.46931(private).there is a huge\r\n gap between this two model in public score,but the 0.46686 model have a little bit higher than 0.47288 model in private score.BTW,I'm also curious what is your best single model.The above two model are both single model,not using any ensemble methods.

", "votes": null, "replies": [] }, { "id": 15788, "author_name": "shinmagi", "author_url": "", "post_date": "10/17/2012 05:54:11", "content": "

my best submitted private score is 0.47704.

", "votes": null, "replies": [] }, { "id": 15792, "author_name": "jasontigg", "author_url": "", "post_date": "10/17/2012 07:56:26", "content": "

I am not in this competition but I noticed that users kaggle rating points have already been modified according to their private scores in this competition, so if you knew what your kaggle score was before and after the competition finished you could reverse\r\n out your position. From the posts on the forum at the time when the score function was being debated it emerged the points are proportional to log10(competitors)/team_size*position^-0.75. I recently won a competition with 80 players in and got around 200 pts.\r\n From that we can see the winner of this competition will have received around 250 pts, 2nd 148pts, 3rd 109 and 4th around 88. So for example i believe Gxav score before this comp ended was around 555 and it is now around 590. That is he has picked up 35 pts\r\n from a team of 3, which scales up to 105 pts, indicating most likely that team DataRobot finished in 3rd place on the private leaderboard. Berry has a score of 155 so he can not have won the private, he must have finished second. Ruslan is part of team gggg\r\n and has a score of 45. He has only entered one competition and that team has 5 members in, so the team score is around 225pt. Ergo Team gggg won.

\r\n

tl;dr I predict private leaderboard is

\r\n

1. gggg

\r\n

2. Berry

\r\n

3. DataRobot

", "votes": null, "replies": [] }, { "id": 15802, "author_name": "ejlok1", "author_url": "", "post_date": "10/17/2012 10:13:21", "content": "

[quote=Jason Tigg;15792]

\r\n

I am not in this competition but I noticed that users kaggle rating points have already been modified according to their private scores in this competition, 

\r\n

[/quote]

\r\n

Wouldn't that just be the public score? In my user profiled, it had assumed the public as the final standing, hence scores calculated from there? no? 

\r\n

In either case, I predict: 

\r\n

1) Data Robot 

\r\n

2) Berry

", "votes": null, "replies": [] }, { "id": 15804, "author_name": "jasontigg", "author_url": "", "post_date": "10/17/2012 10:24:02", "content": "

Its not consistent with public scores as I explained.

", "votes": null, "replies": [] }, { "id": 15806, "author_name": "ccccat", "author_url": "", "post_date": "10/17/2012 11:50:46", "content": "

[quote=Jason Tigg;15792]

\r\n

I am not in this competition but I noticed that users kaggle rating points have already been modified according to their private scores in this competition, so if you knew what your kaggle score was before and after the competition finished you could reverse\r\n out your position. From the posts on the forum at the time when the score function was being debated it emerged the points are proportional to log10(competitors)/team_size*position^-0.75. I recently won a competition with 80 players in and got around 200 pts.\r\n From that we can see the winner of this competition will have received around 250 pts, 2nd 148pts, 3rd 109 and 4th around 88. So for example i believe Gxav score before this comp ended was around 555 and it is now around 590. That is he has picked up 35 pts\r\n from a team of 3, which scales up to 105 pts, indicating most likely that team DataRobot finished in 3rd place on the private leaderboard. Berry has a score of 155 so he can not have won the private, he must have finished second. Ruslan is part of team gggg\r\n and has a score of 45. He has only entered one competition and that team has 5 members in, so the team score is around 225pt. Ergo Team gggg won.

\r\n

tl;dr I predict private leaderboard is

\r\n

1. gggg

\r\n

2. Berry

\r\n

3. DataRobot

\r\n

[/quote]

\r\n

Looks logical, then Grangnam Style is #5 and Luxtorpedia is #6.

\r\n

Small discrepancy is in Berry's total score. He has little bit less points than I would expect taking into account his another competition
\r\n(even with time decay).

\r\n

I cannot reject order:

\r\n

1. gggg

\r\n

2. DataRobot

\r\n

3. Berry

\r\n

 4.

\r\n

5. Grangnam Style

\r\n

6.  Luxtorpedia

\r\n

 

\r\n

 

", "votes": null, "replies": [] }, { "id": 15807, "author_name": "jasontigg", "author_url": "", "post_date": "10/17/2012 11:57:35", "content": "

I believe given their ephemeral nature that hackathons are weighted like 1/3 of normal competitions, which would explain what I will refer to as the "Berry Anomaly"

", "votes": null, "replies": [] }, { "id": 15808, "author_name": "ccccat", "author_url": "", "post_date": "10/17/2012 12:00:52", "content": "

[quote=Jason Tigg;15807]

\r\n

I believe given their ephemeral nature that hackathons are weighted like 1/3 of normal competitions, which would explain what I will refer to as the "Berry Anomaly"

\r\n

[/quote]

\r\n

OK, I probably missed information about that 1/3 coefficient

\r\n

P.S. I found it, per Ben's message in "Users ranking methods?"   hackathrons are weighted as %25

", "votes": null, "replies": [] }, { "id": 15816, "author_name": "rkirana", "author_url": "", "post_date": "10/17/2012 14:42:53", "content": "

sounds logical - given that different folks have different points. I was 144 before this and have moved to 186 which means I got 42 points.
\r\nHowever this is consistent with the public leaderboard rank of 7

", "votes": null, "replies": [] }, { "id": 15817, "author_name": "rkirana", "author_url": "", "post_date": "10/17/2012 14:43:54", "content": "

don't understand why private leaderboard results cannot be published - what is it that is being verified now that could not have been done earlier?

", "votes": null, "replies": [] }, { "id": 15863, "author_name": "ccccat", "author_url": "", "post_date": "10/18/2012 00:33:42", "content": "

"Jason Tigg" data leak was closed. Ranking is back to precompetition level.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "15768": "", "15773": "", "15779": "", "15780": "", "15782": "", "15784": "", "15788": "", "15792": "", "15802": "", "15804": "", "15806": "", "15807": "", "15808": "", "15816": "", "15817": "", "15863": "" }, "source": "meta" }