I am taking Kaggle competitions mainly because they are fun to play in spare time. But usually I don't have the time to make it to the end. So I am wondering if it is OK to discuss our solutions before competitions end - this may give other participants\r\n some isights to make their own solutions even better.
\r\nI have read the rules under the section "It's OK to share code or data if made available to all players, such as on the forums.". But I am not sure if it is OK to do so in the middle of a competition. I will appreciate it if someone (probably from\r\n Kaggle) can explain a little bit more on this.
\r\nThanks!
", "messages": [ { "id": "20264", "postDate": "02/10/2013 12:37:11", "content": "I am taking Kaggle competitions mainly because they are fun to play in spare time. But usually I don't have the time to make it to the end. So I am wondering if it is OK to discuss our solutions before competitions end - this may give other participants\r\n some isights to make their own solutions even better.
\r\nI have read the rules under the section "It's OK to share code or data if made available to all players, such as on the forums.". But I am not sure if it is OK to do so in the middle of a competition. I will appreciate it if someone (probably from\r\n Kaggle) can explain a little bit more on this.
\r\nThanks!
", "rawMarkdown": "", "votes": null }, { "id": "20274", "postDate": "02/10/2013 17:48:39", "content": "Yes. It's OK.
", "rawMarkdown": "", "votes": null }, { "id": "20284", "postDate": "02/10/2013 23:58:55", "content": "Thanks Hrishikesh Huilgolkar.
\r\n\r\n
So this is my solution so far. Be warned it could be a spoiler. I hope someone find it useful. Even better, if someone can play with it and point to me how to improve it further, pretty much like what we do in\r\ncode review. : ) Good luck and have fun.
\r\n\r\n
http://datathinking.wordpress.com/2013/02/10/event-recommendation-engine-challenge-kaggle/
", "rawMarkdown": "", "votes": null }, { "id": "20299", "postDate": "02/11/2013 04:16:38", "content": "Nice! Watch out my blog too at http://blogicious.com I will publish my solution soon.
\r\nIts not as good as yours.. but whatever works :)
\r\n", "rawMarkdown": "", "votes": null }, { "id": "20350", "postDate": "02/12/2013 04:19:36", "content": "
Hi Dolaameng,
\r\nHow important did you find the community feature to be in your models? I have a similar feature set / target function but can't seem to get above 0.65.
\r\nCheers.
", "rawMarkdown": "", "votes": null }, { "id": "20353", "postDate": "02/12/2013 05:23:42", "content": "Hi Saeh,
\r\nI have tried different feature ranking methods for this problem. Among them, randomForest consistenly ranked the "community" feature as important, along with others such as "notification_ahead_hours", "invited" and "event_topic". A glm model gave different\r\n preferences of those features because it is known to be more sensitive to the correlations among features, but again "community" can be found in its top feature list.
\r\nI have also tried to cluster users based on other ideas, such as by their locale, gender, events_of_interest and friendships, but it turns out that those clustering results are either to expensive to compute or not performing as well as the one based on\r\n community detection. I had to limit the numbers of clusters for users/events within 32, since I was using randomForest package in R.
\r\nMy experience was that picking a good target function can go a long way. And since that information is not directly available in the training set directly, my strategy was basicly trial-and-error.
\r\nGood luck!
", "rawMarkdown": "", "votes": null }, { "id": "20434", "postDate": "02/13/2013 18:33:06", "content": "Here's a free tip - if you haven't filtered the duplicated rows, do so now. It improve my score quite a bit, and I was stupid to ignore them till the "last day".
", "rawMarkdown": "", "votes": null }, { "id": "20516", "postDate": "02/16/2013 07:07:29", "content": "FWIW, here is my solution:
\r\nhttp://sujitpal.blogspot.com/2013/02/my-solution-to-kaggle-event.html
\r\nI'm trying to get some experience doing this, would appreciate pointers on how to improve if possible. Doesn't have to be before the challenge ends though, I am probably not going to try and improve the solution for the challenge.
\r\nThanks,
\r\nSujit
", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 20274, "author_name": "", "author_url": "", "post_date": "02/10/2013 17:48:39", "content": "Yes. It's OK.
", "votes": null, "replies": [] }, { "id": 20284, "author_name": "truedolaameng", "author_url": "", "post_date": "02/10/2013 23:58:55", "content": "Thanks Hrishikesh Huilgolkar.
\r\n\r\n
So this is my solution so far. Be warned it could be a spoiler. I hope someone find it useful. Even better, if someone can play with it and point to me how to improve it further, pretty much like what we do in\r\ncode review. : ) Good luck and have fun.
\r\n\r\n
http://datathinking.wordpress.com/2013/02/10/event-recommendation-engine-challenge-kaggle/
", "votes": null, "replies": [] }, { "id": 20299, "author_name": "", "author_url": "", "post_date": "02/11/2013 04:16:38", "content": "Nice! Watch out my blog too at http://blogicious.com I will publish my solution soon.
\r\nIts not as good as yours.. but whatever works :)
\r\n", "votes": null, "replies": [] }, { "id": 20350, "author_name": "thomasatkins", "author_url": "", "post_date": "02/12/2013 04:19:36", "content": "
Hi Dolaameng,
\r\nHow important did you find the community feature to be in your models? I have a similar feature set / target function but can't seem to get above 0.65.
\r\nCheers.
", "votes": null, "replies": [] }, { "id": 20353, "author_name": "truedolaameng", "author_url": "", "post_date": "02/12/2013 05:23:42", "content": "Hi Saeh,
\r\nI have tried different feature ranking methods for this problem. Among them, randomForest consistenly ranked the "community" feature as important, along with others such as "notification_ahead_hours", "invited" and "event_topic". A glm model gave different\r\n preferences of those features because it is known to be more sensitive to the correlations among features, but again "community" can be found in its top feature list.
\r\nI have also tried to cluster users based on other ideas, such as by their locale, gender, events_of_interest and friendships, but it turns out that those clustering results are either to expensive to compute or not performing as well as the one based on\r\n community detection. I had to limit the numbers of clusters for users/events within 32, since I was using randomForest package in R.
\r\nMy experience was that picking a good target function can go a long way. And since that information is not directly available in the training set directly, my strategy was basicly trial-and-error.
\r\nGood luck!
", "votes": null, "replies": [] }, { "id": 20434, "author_name": "roulinir", "author_url": "", "post_date": "02/13/2013 18:33:06", "content": "Here's a free tip - if you haven't filtered the duplicated rows, do so now. It improve my score quite a bit, and I was stupid to ignore them till the "last day".
", "votes": null, "replies": [] }, { "id": 20516, "author_name": "sujitpal", "author_url": "", "post_date": "02/16/2013 07:07:29", "content": "FWIW, here is my solution:
\r\nhttp://sujitpal.blogspot.com/2013/02/my-solution-to-kaggle-event.html
\r\nI'm trying to get some experience doing this, would appreciate pointers on how to improve if possible. Doesn't have to be before the challenge ends though, I am probably not going to try and improve the solution for the challenge.
\r\nThanks,
\r\nSujit
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "20264": "", "20274": "", "20284": "", "20299": "", "20350": "", "20353": "", "20434": "", "20516": "" }, "source": "meta" }