{ "id": 6283, "title": "whats up with MG133*?", "url": "/competitions/accelerometer-biometric-competition/discussion/6283", "author_name": "", "post_date": "2013-11-11T15:13:33.740Z", "votes": null, "comment_count": 7, "views": 2885, "content": "

As many as 23 users in Kaggle with usernames MG133* and 10 in this competition. Can any MG133* explain if they are from a university or something? It looks so weird!

", "messages": [ { "id": "33563", "postDate": "11/11/2013 15:13:33", "content": "

As many as 23 users in Kaggle with usernames MG133* and 10 in this competition. Can any MG133* explain if they are from a university or something? It looks so weird!

", "rawMarkdown": "", "votes": null }, { "id": "33567", "postDate": "11/11/2013 16:36:20", "content": "

Just a class project, like cs725*.

", "rawMarkdown": "", "votes": null }, { "id": "33610", "postDate": "11/12/2013 11:51:12", "content": "

Data  Mining  Assignment

", "rawMarkdown": "", "votes": null }, { "id": "34921", "postDate": "11/19/2013 18:11:02", "content": "

Well, 10 out of the top 50 entries are now MG's.

\n

This is a very talented class. I just hope they play by the rules and do not share code.

\n

This competition seems a bad choice though for a class project, with data that contains leakage.

\n

 

", "rawMarkdown": "", "votes": null }, { "id": "34933", "postDate": "11/19/2013 20:36:37", "content": "

I agree, it looks like a very "talented" class.  While I can imagine that they have the advantage of using the documented data leaks from the forums, it should still take a bunch of work to squeeze the value out of those leaks.  Of course I am mostly concerned about the three MG133*'s ahead of me. ;) 

\n

On the other hand, you still learn a lot even with the data leaks.  The scores may not be as high in a real world problem, but the process is the same.

", "rawMarkdown": "", "votes": null }, { "id": "34942", "postDate": "11/20/2013 00:01:35", "content": "

I believe Menno's concern was more about sharing code between multiple entries, rather than exploiting leaks. The latter is deemed legal anyway, for this competition in particular.

\n

The former, however, is still illegal (I believe). If you'll be sharing code, you must form a team and submit as such. I trust Kaggle's filters on catching rule breaks in that venue.

\n

I'm also really curious about the final standings of the .99 scores. Public LB is 30% of test set, so they won't differ hugely. But still, I think they might be overfitting the LB, so some slight shuffles may occur on the private LB.

\n

Anyways, looking forward to the leak-free version of this competition.

\n

 

", "rawMarkdown": "", "votes": null }, { "id": "34948", "postDate": "11/20/2013 01:19:40", "content": "

[quote=geringer;34933]

\n

On the other hand, you still learn a lot even with the data leaks.

\n

[/quote]

\n

I think you learn a lot because of the leaks. Maybe you don't learn what you expected to learn from a competition of this sort, but you can learn a lot about data analysis from its flaws. Not to mention that you also learn what potential problems to look for if you're preparing data for a machine learning task.

\n

 

", "rawMarkdown": "", "votes": null }, { "id": "34962", "postDate": "11/20/2013 12:10:15", "content": "

I totally agree with you José. Should have used “odd” instead of “bad” :)

\n

Wish William all the best with cleaning up the leaderboard after the competition. It is a big mess out there with double profiles.

\n

I am also looking forward to the results on the private LB just as barisumog. One note though, I believe the private LB results are based on  "the other 70%", so excluding the test data which determines the public LB. This might make it even more interesting...

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 33567, "author_name": "knightchan", "author_url": "", "post_date": "11/11/2013 16:36:20", "content": "

Just a class project, like cs725*.

", "votes": null, "replies": [] }, { "id": 33610, "author_name": "zizhuo0", "author_url": "", "post_date": "11/12/2013 11:51:12", "content": "

Data  Mining  Assignment

", "votes": null, "replies": [] }, { "id": 34921, "author_name": "mennotaanman", "author_url": "", "post_date": "11/19/2013 18:11:02", "content": "

Well, 10 out of the top 50 entries are now MG's.

\n

This is a very talented class. I just hope they play by the rules and do not share code.

\n

This competition seems a bad choice though for a class project, with data that contains leakage.

\n

 

", "votes": null, "replies": [] }, { "id": 34933, "author_name": "geringer", "author_url": "", "post_date": "11/19/2013 20:36:37", "content": "

I agree, it looks like a very "talented" class.  While I can imagine that they have the advantage of using the documented data leaks from the forums, it should still take a bunch of work to squeeze the value out of those leaks.  Of course I am mostly concerned about the three MG133*'s ahead of me. ;) 

\n

On the other hand, you still learn a lot even with the data leaks.  The scores may not be as high in a real world problem, but the process is the same.

", "votes": null, "replies": [] }, { "id": 34942, "author_name": "barisumog", "author_url": "", "post_date": "11/20/2013 00:01:35", "content": "

I believe Menno's concern was more about sharing code between multiple entries, rather than exploiting leaks. The latter is deemed legal anyway, for this competition in particular.

\n

The former, however, is still illegal (I believe). If you'll be sharing code, you must form a team and submit as such. I trust Kaggle's filters on catching rule breaks in that venue.

\n

I'm also really curious about the final standings of the .99 scores. Public LB is 30% of test set, so they won't differ hugely. But still, I think they might be overfitting the LB, so some slight shuffles may occur on the private LB.

\n

Anyways, looking forward to the leak-free version of this competition.

\n

 

", "votes": null, "replies": [] }, { "id": 34948, "author_name": "solorzano", "author_url": "", "post_date": "11/20/2013 01:19:40", "content": "

[quote=geringer;34933]

\n

On the other hand, you still learn a lot even with the data leaks.

\n

[/quote]

\n

I think you learn a lot because of the leaks. Maybe you don't learn what you expected to learn from a competition of this sort, but you can learn a lot about data analysis from its flaws. Not to mention that you also learn what potential problems to look for if you're preparing data for a machine learning task.

\n

 

", "votes": null, "replies": [] }, { "id": 34962, "author_name": "mennotaanman", "author_url": "", "post_date": "11/20/2013 12:10:15", "content": "

I totally agree with you José. Should have used “odd” instead of “bad” :)

\n

Wish William all the best with cleaning up the leaderboard after the competition. It is a big mess out there with double profiles.

\n

I am also looking forward to the results on the private LB just as barisumog. One note though, I believe the private LB results are based on  "the other 70%", so excluding the test data which determines the public LB. This might make it even more interesting...

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "33563": "", "33567": "", "33610": "", "34921": "", "34933": "", "34942": "", "34948": "", "34962": "" }, "source": "meta" }