{ "id": 13229, "title": "tgtg ...", "url": "/competitions/malware-classification/discussion/13229", "author_name": "", "post_date": "2015-04-03T23:16:37.593Z", "votes": 12, "comment_count": 20, "views": 4709, "content": "

So this guy joined 2 hours ago... and is already 2nd in the leader board? Foolish move

", "messages": [ { "id": "69576", "postDate": "04/03/2015 23:16:37", "content": "

So this guy joined 2 hours ago... and is already 2nd in the leader board? Foolish move

", "rawMarkdown": "", "votes": null }, { "id": "69578", "postDate": "04/03/2015 23:22:39", "content": "

It is not necessary to be a multiple account thing. But just downloading and extracting data will take an hour for me... maybe he or she gets data from a friend :-o

", "rawMarkdown": "", "votes": null }, { "id": "69580", "postDate": "04/03/2015 23:25:30", "content": "

yeah...right :/ I don't think so!

", "rawMarkdown": "", "votes": null }, { "id": "69583", "postDate": "04/03/2015 23:37:51", "content": "

I'm just speculating. But it really doesn't have many logical explanations. In my opinion the most likely thing is one of laks_man's labmates decide to show up after a few days' work. I apologize to laks_man and tgtg if you find this offensive.  

", "rawMarkdown": "", "votes": null }, { "id": "69584", "postDate": "04/03/2015 23:41:21", "content": "

I don't know who tgtg is.

", "rawMarkdown": "", "votes": null }, { "id": "69585", "postDate": "04/03/2015 23:42:53", "content": "

[quote=Lakshman Nataraj;69584]

\n

I don't know who tgtg is.

\n

[/quote]

\n

I am sorry. I'm just saying tgtg has to get data before he or she creates the account.

", "rawMarkdown": "", "votes": null }, { "id": "69586", "postDate": "04/03/2015 23:53:59", "content": "

I challenge tgtg that joined kaggle for the first time 2 hours ago and downloaded this...data within 2 hours and had the time (with no collaboration of other players) to make a top 2 submission to reveal how he did it after the competition ends. 

\n

In any case I will post the screen shot as it looks really spooky to me! Hopefully its validity will be investigated. 

", "rawMarkdown": "", "votes": null }, { "id": "69587", "postDate": "04/04/2015 00:06:11", "content": "

no matter what is happening, this is really amazing performance..

\n

0.0030 now..

", "rawMarkdown": "", "votes": null }, { "id": "69590", "postDate": "04/04/2015 00:15:34", "content": "

infinite submissions limit! coolio

\n

[Edit] Bah! forgot that I'm in a different timezone than usual today :)

", "rawMarkdown": "", "votes": null }, { "id": "69591", "postDate": "04/04/2015 01:09:18", "content": "

This is Abhishek's machine that, after so many Kaggle competitions, gained self consciousness, went on Kaggle, created a profile, and started making submissions.

\n

I told you guys that it was only a matter of time before robots would take over Kaggle...

\n

Next we will see "Beat the Benchmark with 0 log loss in 2 minutes and less than a MB of RAM"...

", "rawMarkdown": "", "votes": null }, { "id": "69592", "postDate": "04/04/2015 01:20:00", "content": "

A lot of data points like ting and zhengquexing look suspicious but at this point Kagglers don't have enough data to show whether such users are duplicates of already existing accounts (e.g. cheaters) or just lucky new users. This shouldn't discourage Kagglers from doing an analysis of submissions. I believe a few Kagglers did some very detailed analysis to root out multiple accounts based upon submission time correlations.

", "rawMarkdown": "", "votes": null }, { "id": "69594", "postDate": "04/04/2015 01:25:16", "content": "

Many things strange in LB... last day there are two teams with the same score:

\n

162494 ting 2015-04-02 21:49:39 0.0087485

\n

161940 guasars 2015-04-02 14:23:10 0.0087485

\n\n

Now this fantastic tgtg performance, just 2 hous after joined kaggle:

\n

162852 tgtg 2015-04-03 23:49:29 0.0036089
162852 tgtg 2015-04-04 00:03:43 0.0030238
147873 laks_man 2015-04-04 00:06:15 0.0027026
162852 tgtg 2015-04-04 00:10:30 0.0021241

\n\n

Congrats tgtg, I'm eager to know your solution!!!

\n

Admins will have work after this one...

", "rawMarkdown": "", "votes": null }, { "id": "69596", "postDate": "04/04/2015 01:45:00", "content": "

This looks tricky as hell, I do not see anything wrong with tgtg, I mean, lets analyze it from a logical point of view.

\n

1) To stay under rules, you can not work on a solution without first agree to the rules and download the data.
2) The only way to do step 1 is by getting an account first.
3) I am envy for such internet speed, lets say the guy has a 200 Mbps connection, so he/she/it managed to download the over 35GB at a constant 18 Mb/s in just 34 minutes.
4) Since account was made 2 hours before submission, then you have less than 1.5 hrs to 1) Extract 35GB, 2) Code your feature extraction, 3) Run your feature extraction, 4) Training your model, 5) Fine tune your model, 6) Ensemble stuff and upload prediction

\n

I do not know you guys but maybe you are just too lazy or too slow, but 1.5 hrs sounds like plenty of time to do that, I mean, I reached 0.007 in less than 38.67 seconds with my 1000-CPU cluster at home, and this includes me typing at light speed my code.

", "rawMarkdown": "", "votes": null }, { "id": "69601", "postDate": "04/04/2015 03:56:40", "content": "

I need to apologize to Lakshman again. Now it is clear.

", "rawMarkdown": "", "votes": null }, { "id": "69602", "postDate": "04/04/2015 04:10:04", "content": "

Looks like tgtg just joined a team... and improved the score to 0.0010098. Time to make some popcorn.

", "rawMarkdown": "", "votes": null }, { "id": "69603", "postDate": "04/04/2015 04:25:10", "content": "

[quote=rcarson;69601]

\n

I need to apologize to Lakshman again. Now it is clear.

\n

[/quote]

\n

That's ok.

\n

It's good that they have now formed a team and probably have a pretty good technique at hand.

\n

But the important point is that they were submitting more number of submissions as individuals before they formed the team and they had probably shared the same data. The number of submissions per day is very crucial. Else, I can easily post many submissions by sharing my code/data with others and try to keep improving my technique but I've only been doing what I'm allowed to do.  I hope the admins take note of it.

", "rawMarkdown": "", "votes": null }, { "id": "69604", "postDate": "04/04/2015 04:43:17", "content": "

[quote=Lakshman Nataraj;69603]

\n

[quote=rcarson;69601]

\n

I need to apologize to Lakshman again. Now it is clear.

\n

[/quote]

\n

That's ok.

\n

It's good that they have now formed a team and probably have a pretty good technique at hand.

\n

But the important point is that they were submitting more number of submissions as individuals before they formed the team and they had probably shared the same data. The number of submissions per day is very crucial. Else, I can easily post many submissions by sharing my code/data with others and try to keep improving my technique but I've only been doing what I'm allowed to do.  I hope the admins take note of it.

\n

[/quote]

\n

Thank you.

\n

I don't worry any more now. Kaggle has been enforcing the rule very strictly and this is much wilder than the case removed in BCI and American Seizure Prediction. There is at least private data sharing to tgtg now before forming a team and that is against the rule.  

", "rawMarkdown": "", "votes": null }, { "id": "69723", "postDate": "04/05/2015 17:34:47", "content": "

The guy must have absurd computational and download speed and here am I with 8GB RAM Laptop and 10Mbps download speed.

", "rawMarkdown": "", "votes": null }, { "id": "69810", "postDate": "04/06/2015 16:43:41", "content": "

Was this suspicious paper written by the 1st ranked team's members (SSIR) ?

\n

http://users.ece.cmu.edu/~jiyongj/papers/ccs11.pdf

", "rawMarkdown": "", "votes": null }, { "id": "69873", "postDate": "04/07/2015 11:31:05", "content": "

Why is it suspicious?

", "rawMarkdown": "", "votes": null }, { "id": "69915", "postDate": "04/07/2015 18:30:07", "content": "

Mikhail, I was wondering if they used the technique in the paper to achieve a very good score. :)

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 69578, "author_name": "jiweiliu", "author_url": "", "post_date": "04/03/2015 23:22:39", "content": "

It is not necessary to be a multiple account thing. But just downloading and extracting data will take an hour for me... maybe he or she gets data from a friend :-o

", "votes": null, "replies": [] }, { "id": 69580, "author_name": "kazanova", "author_url": "", "post_date": "04/03/2015 23:25:30", "content": "

yeah...right :/ I don't think so!

", "votes": null, "replies": [] }, { "id": 69583, "author_name": "jiweiliu", "author_url": "", "post_date": "04/03/2015 23:37:51", "content": "

I'm just speculating. But it really doesn't have many logical explanations. In my opinion the most likely thing is one of laks_man's labmates decide to show up after a few days' work. I apologize to laks_man and tgtg if you find this offensive.  

", "votes": null, "replies": [] }, { "id": 69584, "author_name": "", "author_url": "", "post_date": "04/03/2015 23:41:21", "content": "

I don't know who tgtg is.

", "votes": null, "replies": [] }, { "id": 69585, "author_name": "jiweiliu", "author_url": "", "post_date": "04/03/2015 23:42:53", "content": "

[quote=Lakshman Nataraj;69584]

\n

I don't know who tgtg is.

\n

[/quote]

\n

I am sorry. I'm just saying tgtg has to get data before he or she creates the account.

", "votes": null, "replies": [] }, { "id": 69586, "author_name": "kazanova", "author_url": "", "post_date": "04/03/2015 23:53:59", "content": "

I challenge tgtg that joined kaggle for the first time 2 hours ago and downloaded this...data within 2 hours and had the time (with no collaboration of other players) to make a top 2 submission to reveal how he did it after the competition ends. 

\n

In any case I will post the screen shot as it looks really spooky to me! Hopefully its validity will be investigated. 

", "votes": null, "replies": [] }, { "id": 69587, "author_name": "jiweiliu", "author_url": "", "post_date": "04/04/2015 00:06:11", "content": "

no matter what is happening, this is really amazing performance..

\n

0.0030 now..

", "votes": null, "replies": [] }, { "id": 69590, "author_name": "ashhafez", "author_url": "", "post_date": "04/04/2015 00:15:34", "content": "

infinite submissions limit! coolio

\n

[Edit] Bah! forgot that I'm in a different timezone than usual today :)

", "votes": null, "replies": [] }, { "id": 69591, "author_name": "adjgiulio", "author_url": "", "post_date": "04/04/2015 01:09:18", "content": "

This is Abhishek's machine that, after so many Kaggle competitions, gained self consciousness, went on Kaggle, created a profile, and started making submissions.

\n

I told you guys that it was only a matter of time before robots would take over Kaggle...

\n

Next we will see "Beat the Benchmark with 0 log loss in 2 minutes and less than a MB of RAM"...

", "votes": null, "replies": [] }, { "id": 69592, "author_name": "mikeskim", "author_url": "", "post_date": "04/04/2015 01:20:00", "content": "

A lot of data points like ting and zhengquexing look suspicious but at this point Kagglers don't have enough data to show whether such users are duplicates of already existing accounts (e.g. cheaters) or just lucky new users. This shouldn't discourage Kagglers from doing an analysis of submissions. I believe a few Kagglers did some very detailed analysis to root out multiple accounts based upon submission time correlations.

", "votes": null, "replies": [] }, { "id": 69594, "author_name": "titericz", "author_url": "", "post_date": "04/04/2015 01:25:16", "content": "

Many things strange in LB... last day there are two teams with the same score:

\n

162494 ting 2015-04-02 21:49:39 0.0087485

\n

161940 guasars 2015-04-02 14:23:10 0.0087485

\n\n

Now this fantastic tgtg performance, just 2 hous after joined kaggle:

\n

162852 tgtg 2015-04-03 23:49:29 0.0036089
162852 tgtg 2015-04-04 00:03:43 0.0030238
147873 laks_man 2015-04-04 00:06:15 0.0027026
162852 tgtg 2015-04-04 00:10:30 0.0021241

\n\n

Congrats tgtg, I'm eager to know your solution!!!

\n

Admins will have work after this one...

", "votes": null, "replies": [] }, { "id": 69596, "author_name": "carloshuertas", "author_url": "", "post_date": "04/04/2015 01:45:00", "content": "

This looks tricky as hell, I do not see anything wrong with tgtg, I mean, lets analyze it from a logical point of view.

\n

1) To stay under rules, you can not work on a solution without first agree to the rules and download the data.
2) The only way to do step 1 is by getting an account first.
3) I am envy for such internet speed, lets say the guy has a 200 Mbps connection, so he/she/it managed to download the over 35GB at a constant 18 Mb/s in just 34 minutes.
4) Since account was made 2 hours before submission, then you have less than 1.5 hrs to 1) Extract 35GB, 2) Code your feature extraction, 3) Run your feature extraction, 4) Training your model, 5) Fine tune your model, 6) Ensemble stuff and upload prediction

\n

I do not know you guys but maybe you are just too lazy or too slow, but 1.5 hrs sounds like plenty of time to do that, I mean, I reached 0.007 in less than 38.67 seconds with my 1000-CPU cluster at home, and this includes me typing at light speed my code.

", "votes": null, "replies": [] }, { "id": 69601, "author_name": "jiweiliu", "author_url": "", "post_date": "04/04/2015 03:56:40", "content": "

I need to apologize to Lakshman again. Now it is clear.

", "votes": null, "replies": [] }, { "id": 69602, "author_name": "davidtran", "author_url": "", "post_date": "04/04/2015 04:10:04", "content": "

Looks like tgtg just joined a team... and improved the score to 0.0010098. Time to make some popcorn.

", "votes": null, "replies": [] }, { "id": 69603, "author_name": "", "author_url": "", "post_date": "04/04/2015 04:25:10", "content": "

[quote=rcarson;69601]

\n

I need to apologize to Lakshman again. Now it is clear.

\n

[/quote]

\n

That's ok.

\n

It's good that they have now formed a team and probably have a pretty good technique at hand.

\n

But the important point is that they were submitting more number of submissions as individuals before they formed the team and they had probably shared the same data. The number of submissions per day is very crucial. Else, I can easily post many submissions by sharing my code/data with others and try to keep improving my technique but I've only been doing what I'm allowed to do.  I hope the admins take note of it.

", "votes": null, "replies": [] }, { "id": 69604, "author_name": "jiweiliu", "author_url": "", "post_date": "04/04/2015 04:43:17", "content": "

[quote=Lakshman Nataraj;69603]

\n

[quote=rcarson;69601]

\n

I need to apologize to Lakshman again. Now it is clear.

\n

[/quote]

\n

That's ok.

\n

It's good that they have now formed a team and probably have a pretty good technique at hand.

\n

But the important point is that they were submitting more number of submissions as individuals before they formed the team and they had probably shared the same data. The number of submissions per day is very crucial. Else, I can easily post many submissions by sharing my code/data with others and try to keep improving my technique but I've only been doing what I'm allowed to do.  I hope the admins take note of it.

\n

[/quote]

\n

Thank you.

\n

I don't worry any more now. Kaggle has been enforcing the rule very strictly and this is much wilder than the case removed in BCI and American Seizure Prediction. There is at least private data sharing to tgtg now before forming a team and that is against the rule.  

", "votes": null, "replies": [] }, { "id": 69723, "author_name": "jbsilva", "author_url": "", "post_date": "04/05/2015 17:34:47", "content": "

The guy must have absurd computational and download speed and here am I with 8GB RAM Laptop and 10Mbps download speed.

", "votes": null, "replies": [] }, { "id": 69810, "author_name": "mahmadi", "author_url": "", "post_date": "04/06/2015 16:43:41", "content": "

Was this suspicious paper written by the 1st ranked team's members (SSIR) ?

\n

http://users.ece.cmu.edu/~jiyongj/papers/ccs11.pdf

", "votes": null, "replies": [] }, { "id": 69873, "author_name": "mikhailtrofimov", "author_url": "", "post_date": "04/07/2015 11:31:05", "content": "

Why is it suspicious?

", "votes": null, "replies": [] }, { "id": 69915, "author_name": "mahmadi", "author_url": "", "post_date": "04/07/2015 18:30:07", "content": "

Mikhail, I was wondering if they used the technique in the paper to achieve a very good score. :)

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "69576": "", "69578": "", "69580": "", "69583": "", "69584": "", "69585": "", "69586": "", "69587": "", "69590": "", "69591": "", "69592": "", "69594": "", "69596": "", "69601": "", "69602": "", "69603": "", "69604": "", "69723": "", "69810": "", "69873": "", "69915": "" }, "source": "meta" }