{ "id": 14468, "title": "Ideas sharing", "url": "/competitions/facebook-recruiting-iv-human-or-bot/discussion/14468", "author_name": "", "post_date": "2015-06-01T09:53:49.320Z", "votes": 3, "comment_count": 11, "views": 2272, "content": "

Hello, all!

\n

Competition is ending soon.

\n

I want to share some ideas and ask some questions about features engineering and model tuning.

\n

However it's not clear for me that it's ok to do that. There are nothing about concrete ideas sharing in rules. Rules saying only about code sharing. However some of the ideas could be as useful as a piece of the code (at the end it's not a programming contest).

\n

I think I need some clarification from the competition admins before proceeding.

", "messages": [ { "id": "80563", "postDate": "06/01/2015 09:53:49", "content": "

Hello, all!

\n

Competition is ending soon.

\n

I want to share some ideas and ask some questions about features engineering and model tuning.

\n

However it's not clear for me that it's ok to do that. There are nothing about concrete ideas sharing in rules. Rules saying only about code sharing. However some of the ideas could be as useful as a piece of the code (at the end it's not a programming contest).

\n

I think I need some clarification from the competition admins before proceeding.

", "rawMarkdown": "", "votes": null }, { "id": "80583", "postDate": "06/01/2015 15:44:21", "content": "

The competition just ends in less than 8 days, there is no hurry, I am pretty sure that after the competition ends, there will be lots of ideas/code sharing.

\n

You could focus on a last push by now.

", "rawMarkdown": "", "votes": null }, { "id": "80590", "postDate": "06/01/2015 17:30:39", "content": "

last push, I just started...

", "rawMarkdown": "", "votes": null }, { "id": "80608", "postDate": "06/01/2015 19:57:47", "content": "

I have an interest to communicate before competition end. So I would like to hear comments from competition admin.

", "rawMarkdown": "", "votes": null }, { "id": "80610", "postDate": "06/01/2015 20:01:10", "content": "

Please dont until after the competition. I dont necessarily think that it matters in any real sense.

\n

However, no matter how useful or not any idea you share in this one thread people will make many more threads using this thread as a scapegoat for any change in their leaderboard position.

\n

It happens every time in the public competitions that a benchmark or anything is released. It is a human trait to seek a scapegoat and this forum has been fairly on topic and productive.

", "rawMarkdown": "", "votes": null }, { "id": "80616", "postDate": "06/01/2015 20:51:05", "content": "

Guys I'm understanding that somebody don't like that people share ideas or code in competitions.

\n

But in the long distance all the community wins because while people copying they are learning something. Today they will be at top-100 because they copied some. But tomorrow same people would be in top-10 generating their own ideas worth copying.

\n

After all this competition is not about money or any benefits. If you really want to be at FB you could have interview with them without competition. If you can't get to interview point you probably couldn't pass it.

\n

So it's only about knowledge and fun. And for both things it's ok to share (if it's not against rules).

\n

For the person who set minus for my previous post:

\n

I hope you understand that it's more about rules -- I ask about rules and I would follow them. If you don't like rules of competitions where it's ok to share solutions during contest it's better to skip them, same for ideas part. In this sport (Kaggle competitions) power of copying good solutions is about the same like building your own. You accept this rule and by your own will you are part of this sport.

", "rawMarkdown": "", "votes": null }, { "id": "80619", "postDate": "06/01/2015 21:12:35", "content": "

Aside from the reasons I already gave I don't want this to actually happen because It wouldnt really add too many ideas. Nobody is going to give their best ideas only their throwaway obvious ideas yet many people will assume that implementing those ideas is the secret sauce  that caused them to drop in the leaderboard.

\n

Other threads in this forum already give ideas. If you look at the other threads you should be able to derive ideas from the threads easily and in some cases it is explicitly spelled out.

\n

The only thing making a a thread called "idea sharing" and sharing ideas does is make it easier for the people looking to scapegoat their performance because they can point to it have others look at the thread title and make it easier for the bandwagon of victimization to form. This is especially true in the last week because some people are holding out their better models until this week so the leaderboard will change and people will look for weak correlations to attribute causal blame to. The worst part is that the flame wont even stay contained in this thread but spread to the other more general kaggle forums.

", "rawMarkdown": "", "votes": null }, { "id": "80620", "postDate": "06/01/2015 21:22:56", "content": "

Other threads were not really about ideas, they were rather about how to understand the data – in my opinion some people put too much effort in that. I think Dimitry thinks rather about some modelling ideas, feature selection algorithms, etc. and not whether bid_id denotes bid order or is it only identifier.

", "rawMarkdown": "", "votes": null }, { "id": "80621", "postDate": "06/01/2015 21:29:38", "content": "

[quote=mpekalski;80620]

\n

Other threads were not really about ideas, they were rather about how to understand the data – in my opinion some people put too much effort in that. I think Dimitry thinks rather about some modelling ideas, feature selection algorithms, etc. and not whether bid_id denotes bid order or is it only identifier.

\n

[/quote]

\n

So the multiple threads about time obfuscation and determining if they are still rankable in no way suggests incorporating time  in your model assuming it is ordered.

\n

In more than one instance I have thought "that might be a good feature" based on post on a thread in this forum even if that feature didnt pan out but the idea still has come from the forum.

\n

Make individual threads based on an idea just dont gather them into a single thread and name it "idea sharing" so that it makes for an easy scapegoat.

", "rawMarkdown": "", "votes": null }, { "id": "80622", "postDate": "06/01/2015 21:54:11", "content": "

Basically I want to prevent one thread like:

\n

https://www.kaggle.com/c/otto-group-product-classification-challenge/forums/t/13632/achieve-0-48-in-5-min-with-a-deep-net-feat-batchnorm-prelu

\n

end up cluttering all of Kaggle with threads like:

\n

https://www.kaggle.com/c/otto-group-product-classification-challenge/forums/t/14233/why?page=12

\n

https://www.kaggle.com/forums/f/15/kaggle-forum/t/14143/possible-modification-to-competition-rules-code-sharing

\n

Those people just cooled off and there is no reason to prod them. Sharing ideas in as explicit form as you are suggesting is going to wake them up since you dont need to share actual code to share models . See:

\n

https://www.kaggle.com/c/otto-group-product-classification-challenge/forums/t/14297/share-your-models

\n

Many of those posts are sharing models without a single line of code.

", "rawMarkdown": "", "votes": null }, { "id": "80624", "postDate": "06/01/2015 23:23:17", "content": "

I don't know if my pulling out of FR1 would be classified as "scapegoating" under this definition, but my belief remains that the results of a recruiting competition are easier to untangle if no sharing happens.

", "rawMarkdown": "", "votes": null }, { "id": "80625", "postDate": "06/01/2015 23:56:27", "content": "

[quote=John Costella;80624]

\n

I don't know if my pulling out of FR1 would be classified as "scapegoating" under this definition, but my belief remains that the results of a recruiting competition are easier to untangle if no sharing happens.

\n

[/quote]

\n

Thanks for adding the input as an admin as it helps settle the boundaries for the competition and hopefully puts an end to this thread. It might be useful to somehow add a pinned reminder in the forum otherwise it might end up being needlessly be repeated by admins like the time obfuscation ordering has been.

\n

I just enjoy the lack of drama in the recruiting competitions when compared to the other competitions and I am hoping we can make it a week without someone skirting the rules or making a misstep which will disrupt the peaceful environment of this competition. 

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 80583, "author_name": "carloshuertas", "author_url": "", "post_date": "06/01/2015 15:44:21", "content": "

The competition just ends in less than 8 days, there is no hurry, I am pretty sure that after the competition ends, there will be lots of ideas/code sharing.

\n

You could focus on a last push by now.

", "votes": null, "replies": [] }, { "id": 80590, "author_name": "skylibrary", "author_url": "", "post_date": "06/01/2015 17:30:39", "content": "

last push, I just started...

", "votes": null, "replies": [] }, { "id": 80608, "author_name": "dremovd", "author_url": "", "post_date": "06/01/2015 19:57:47", "content": "

I have an interest to communicate before competition end. So I would like to hear comments from competition admin.

", "votes": null, "replies": [] }, { "id": 80610, "author_name": "jbsilva", "author_url": "", "post_date": "06/01/2015 20:01:10", "content": "

Please dont until after the competition. I dont necessarily think that it matters in any real sense.

\n

However, no matter how useful or not any idea you share in this one thread people will make many more threads using this thread as a scapegoat for any change in their leaderboard position.

\n

It happens every time in the public competitions that a benchmark or anything is released. It is a human trait to seek a scapegoat and this forum has been fairly on topic and productive.

", "votes": null, "replies": [] }, { "id": 80616, "author_name": "dremovd", "author_url": "", "post_date": "06/01/2015 20:51:05", "content": "

Guys I'm understanding that somebody don't like that people share ideas or code in competitions.

\n

But in the long distance all the community wins because while people copying they are learning something. Today they will be at top-100 because they copied some. But tomorrow same people would be in top-10 generating their own ideas worth copying.

\n

After all this competition is not about money or any benefits. If you really want to be at FB you could have interview with them without competition. If you can't get to interview point you probably couldn't pass it.

\n

So it's only about knowledge and fun. And for both things it's ok to share (if it's not against rules).

\n

For the person who set minus for my previous post:

\n

I hope you understand that it's more about rules -- I ask about rules and I would follow them. If you don't like rules of competitions where it's ok to share solutions during contest it's better to skip them, same for ideas part. In this sport (Kaggle competitions) power of copying good solutions is about the same like building your own. You accept this rule and by your own will you are part of this sport.

", "votes": null, "replies": [] }, { "id": 80619, "author_name": "jbsilva", "author_url": "", "post_date": "06/01/2015 21:12:35", "content": "

Aside from the reasons I already gave I don't want this to actually happen because It wouldnt really add too many ideas. Nobody is going to give their best ideas only their throwaway obvious ideas yet many people will assume that implementing those ideas is the secret sauce  that caused them to drop in the leaderboard.

\n

Other threads in this forum already give ideas. If you look at the other threads you should be able to derive ideas from the threads easily and in some cases it is explicitly spelled out.

\n

The only thing making a a thread called "idea sharing" and sharing ideas does is make it easier for the people looking to scapegoat their performance because they can point to it have others look at the thread title and make it easier for the bandwagon of victimization to form. This is especially true in the last week because some people are holding out their better models until this week so the leaderboard will change and people will look for weak correlations to attribute causal blame to. The worst part is that the flame wont even stay contained in this thread but spread to the other more general kaggle forums.

", "votes": null, "replies": [] }, { "id": 80620, "author_name": "mpekalski", "author_url": "", "post_date": "06/01/2015 21:22:56", "content": "

Other threads were not really about ideas, they were rather about how to understand the data – in my opinion some people put too much effort in that. I think Dimitry thinks rather about some modelling ideas, feature selection algorithms, etc. and not whether bid_id denotes bid order or is it only identifier.

", "votes": null, "replies": [] }, { "id": 80621, "author_name": "jbsilva", "author_url": "", "post_date": "06/01/2015 21:29:38", "content": "

[quote=mpekalski;80620]

\n

Other threads were not really about ideas, they were rather about how to understand the data – in my opinion some people put too much effort in that. I think Dimitry thinks rather about some modelling ideas, feature selection algorithms, etc. and not whether bid_id denotes bid order or is it only identifier.

\n

[/quote]

\n

So the multiple threads about time obfuscation and determining if they are still rankable in no way suggests incorporating time  in your model assuming it is ordered.

\n

In more than one instance I have thought "that might be a good feature" based on post on a thread in this forum even if that feature didnt pan out but the idea still has come from the forum.

\n

Make individual threads based on an idea just dont gather them into a single thread and name it "idea sharing" so that it makes for an easy scapegoat.

", "votes": null, "replies": [] }, { "id": 80622, "author_name": "jbsilva", "author_url": "", "post_date": "06/01/2015 21:54:11", "content": "

Basically I want to prevent one thread like:

\n

https://www.kaggle.com/c/otto-group-product-classification-challenge/forums/t/13632/achieve-0-48-in-5-min-with-a-deep-net-feat-batchnorm-prelu

\n

end up cluttering all of Kaggle with threads like:

\n

https://www.kaggle.com/c/otto-group-product-classification-challenge/forums/t/14233/why?page=12

\n

https://www.kaggle.com/forums/f/15/kaggle-forum/t/14143/possible-modification-to-competition-rules-code-sharing

\n

Those people just cooled off and there is no reason to prod them. Sharing ideas in as explicit form as you are suggesting is going to wake them up since you dont need to share actual code to share models . See:

\n

https://www.kaggle.com/c/otto-group-product-classification-challenge/forums/t/14297/share-your-models

\n

Many of those posts are sharing models without a single line of code.

", "votes": null, "replies": [] }, { "id": 80624, "author_name": "costella", "author_url": "", "post_date": "06/01/2015 23:23:17", "content": "

I don't know if my pulling out of FR1 would be classified as "scapegoating" under this definition, but my belief remains that the results of a recruiting competition are easier to untangle if no sharing happens.

", "votes": null, "replies": [] }, { "id": 80625, "author_name": "jbsilva", "author_url": "", "post_date": "06/01/2015 23:56:27", "content": "

[quote=John Costella;80624]

\n

I don't know if my pulling out of FR1 would be classified as "scapegoating" under this definition, but my belief remains that the results of a recruiting competition are easier to untangle if no sharing happens.

\n

[/quote]

\n

Thanks for adding the input as an admin as it helps settle the boundaries for the competition and hopefully puts an end to this thread. It might be useful to somehow add a pinned reminder in the forum otherwise it might end up being needlessly be repeated by admins like the time obfuscation ordering has been.

\n

I just enjoy the lack of drama in the recruiting competitions when compared to the other competitions and I am hoping we can make it a week without someone skirting the rules or making a misstep which will disrupt the peaceful environment of this competition. 

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "80563": "", "80583": "", "80590": "", "80608": "", "80610": "", "80616": "", "80619": "", "80620": "", "80621": "", "80622": "", "80624": "", "80625": "" }, "source": "meta" }