Here is my solution that gives the above score. (I didn't pick this as one of my final submissions !) https://github.com/paragkale/kaggle-airbnb. There are two scripts (1) first run "makedata_session.R" to aggregate the sessions information. (2) then run the other script to get the output. You will get 2 output files - one based on probabilities and one based on ranks. I have used many ideas/code from the public scripts. The trick was to use logistic regression to enhance the splits between the countries and rank averaging.
\n\nIn hindsight, my ensembles let me down. This solution is rather simple(single file) and would place you in second position on the private LB. I had a hard time picking my final submissions - I picked one based on public LB score and one based on my CV scores. I guess I was lucky that I didn't fall out of top10!
", "comments": [ { "id": 107981, "author_name": "Rich He", "author_url": "", "post_date": "2016-02-14T14:17:04.437000", "content": "Thanks pxk. I like your approaching to this competition.
\n\nI ran the model last night and got the same results. So this might be the issue of different settings. I am using Windows 7 professional, R version 3.2.2, xgboost_0.4-2, and Intel Core2 Duo CPU E8400 @3.00GHz.
", "votes": 1, "replies": [] }, { "id": 107965, "author_name": "Parag", "author_url": "", "post_date": "2016-02-14T07:50:22.220000", "content": "Hi Rich, thanks for trying out the script. I resubmitted the output, just to confirm and I get the original results.\nPost-Deadline: Sun, 14 Feb 2016 07:19:00 Edit description sub_u26a_rank.csv 0.88121 0.88691
\n\nReproducibility in a multi-threaded environment has been a challenge and several other people seem to have encountered it, in this competition and few others while using xgboost with multiple threads. For reference, here is my environment (Ubuntu 14.04, R version 3.2.3, xgboost_0.4-2, i5 4core)
", "votes": 0, "replies": [] }, { "id": 107905, "author_name": "Rich He", "author_url": "", "post_date": "2016-02-13T17:13:35.650000", "content": "Thanks for sharing. I ran the code and only got 0.865xx on public, and 0.870xx on private. Are you sure it is the right model? Or I made some mistakes?
\n\nPost-Deadline: Sat, 13 Feb 2016 16:57:11 \nEdit description sub_u26a_prob.csv 0.86582 0.87061
\n\nPost-Deadline: Sat, 13 Feb 2016 16:55:27 \nEdit description sub_u26a_rank.csv 0.86574 0.8709
", "votes": 0, "replies": [] }, { "id": 107875, "author_name": "Abhi", "author_url": "", "post_date": "2016-02-13T11:18:42.873000", "content": "Thanks for the scripts
", "votes": 0, "replies": [] } ], "comment_count": 4, "messages": [ { "id": 107869, "postDate": "2016-02-13T09:46:41.800Z", "content": "Here is my solution that gives the above score. (I didn't pick this as one of my final submissions !) https://github.com/paragkale/kaggle-airbnb. There are two scripts (1) first run "makedata_session.R" to aggregate the sessions information. (2) then run the other script to get the output. You will get 2 output files - one based on probabilities and one based on ranks. I have used many ideas/code from the public scripts. The trick was to use logistic regression to enhance the splits between the countries and rank averaging.
\n\nIn hindsight, my ensembles let me down. This solution is rather simple(single file) and would place you in second position on the private LB. I had a hard time picking my final submissions - I picked one based on public LB score and one based on my CV scores. I guess I was lucky that I didn't fall out of top10!
", "rawMarkdown": "Here is my solution that gives the above score. (I didn't pick this as one of my final submissions !) https://github.com/paragkale/kaggle-airbnb. There are two scripts (1) first run \"makedata_session.R\" to aggregate the sessions information. (2) then run the other script to get the output. You will get 2 output files - one based on probabilities and one based on ranks. I have used many ideas/code from the public scripts. The trick was to use logistic regression to enhance the splits between the countries and rank averaging.\r\n\r\nIn hindsight, my ensembles let me down. This solution is rather simple(single file) and would place you in second position on the private LB. I had a hard time picking my final submissions - I picked one based on public LB score and one based on my CV scores. I guess I was lucky that I didn't fall out of top10!\r\n", "votes": 11 }, { "id": 107981, "postDate": "2016-02-14T14:17:04.437Z", "content": "Thanks pxk. I like your approaching to this competition.
\n\nI ran the model last night and got the same results. So this might be the issue of different settings. I am using Windows 7 professional, R version 3.2.2, xgboost_0.4-2, and Intel Core2 Duo CPU E8400 @3.00GHz.
", "rawMarkdown": "Thanks pxk. I like your approaching to this competition. \r\n\r\nI ran the model last night and got the same results. So this might be the issue of different settings. I am using Windows 7 professional, R version 3.2.2, xgboost_0.4-2, and Intel Core2 Duo CPU E8400 @3.00GHz.\r\n\r\n ", "votes": 1 }, { "id": 107965, "postDate": "2016-02-14T07:50:22.220Z", "content": "Hi Rich, thanks for trying out the script. I resubmitted the output, just to confirm and I get the original results.\nPost-Deadline: Sun, 14 Feb 2016 07:19:00 Edit description sub_u26a_rank.csv 0.88121 0.88691
\n\nReproducibility in a multi-threaded environment has been a challenge and several other people seem to have encountered it, in this competition and few others while using xgboost with multiple threads. For reference, here is my environment (Ubuntu 14.04, R version 3.2.3, xgboost_0.4-2, i5 4core)
", "rawMarkdown": "Hi Rich, thanks for trying out the script. I resubmitted the output, just to confirm and I get the original results.\r\nPost-Deadline: Sun, 14 Feb 2016 07:19:00 Edit description sub_u26a_rank.csv 0.88121 0.88691\r\n\r\nReproducibility in a multi-threaded environment has been a challenge and several other people seem to have encountered it, in this competition and few others while using xgboost with multiple threads. For reference, here is my environment (Ubuntu 14.04, R version 3.2.3, xgboost_0.4-2, i5 4core)" }, { "id": 107905, "postDate": "2016-02-13T17:13:35.650Z", "content": "Thanks for sharing. I ran the code and only got 0.865xx on public, and 0.870xx on private. Are you sure it is the right model? Or I made some mistakes?
\n\nPost-Deadline: Sat, 13 Feb 2016 16:57:11 \nEdit description sub_u26a_prob.csv 0.86582 0.87061
\n\nPost-Deadline: Sat, 13 Feb 2016 16:55:27 \nEdit description sub_u26a_rank.csv 0.86574 0.8709
", "rawMarkdown": "Thanks for sharing. I ran the code and only got 0.865xx on public, and 0.870xx on private. Are you sure it is the right model? Or I made some mistakes?\r\n\r\nPost-Deadline: Sat, 13 Feb 2016 16:57:11 \r\nEdit description sub_u26a_prob.csv 0.86582 0.87061 \r\n\r\nPost-Deadline: Sat, 13 Feb 2016 16:55:27 \r\nEdit description sub_u26a_rank.csv 0.86574 0.8709 " }, { "id": 107875, "postDate": "2016-02-13T11:18:42.873Z", "content": "Thanks for the scripts
", "rawMarkdown": "Thanks for the scripts" } ], "complete": true, "source": "live", "fetched_at": "2026-10-19T13:53:37Z", "competition_id": 4651, "competition_slug": "airbnb-recruiting-new-user-bookings", "team_id": 261758, "solution_url": "/competitions/airbnb-recruiting-new-user-bookings/writeups/pxk-solution-sharing-0-88691-on-private-lb" }