{
  "id": 19098,
  "title": "Code sharing (60th position on private leaderboard)",
  "content": "<p>Hello Everyone,</p>\n\n<p>This was a great competition ! </p>\n\n<p>This is a description of my technique:\n(1)   &quot;Fresh&quot; data got a higher weight.\n An important characteristic of this competition is that there is a time cut-off between training and testing data -- training data end on 7/1/2014, and testing data start on that day. Moreover, sessions are available starting from 1/1/2014. Thus, I used  4 classifiers: (1) XGB trained on all data,  (2) RandomForests trained on all data,  (3) XGB classifier trained on recent (aka fresh) data only, and (4) RandomForests trained on fresh data only. Classifiers' results formed the final prediction via weighted voting, and classifiers (3) and (4) had a higher weight.</p>\n\n<p>(2) I used one-hot encoding for the categorical features in the basic dataset, and simple frequency counts for the sessions data. To avoid overfitting, and speed up computations, I replaced all rare values , e.g. (device_type==Linux Desktop), with &quot;other&quot;.</p>\n\n<p>(3) To tune classifiers and evaluate the prediction I used CV over a simplified NDCG score function.</p>\n\n<p>You can find my code here:\n<a href=\"https://github.com/blerim153/kaggle_airbnb_new_user_bookings\">https://github.com/blerim153/kaggle_airbnb_new_user_bookings</a></p>",
  "comments": [],
  "comment_count": 0,
  "messages": [
    {
      "id": 108877,
      "postDate": "2016-02-21T01:08:07.440Z",
      "content": "<p>Hello Everyone,</p>\n\n<p>This was a great competition ! </p>\n\n<p>This is a description of my technique:\n(1)   &quot;Fresh&quot; data got a higher weight.\n An important characteristic of this competition is that there is a time cut-off between training and testing data -- training data end on 7/1/2014, and testing data start on that day. Moreover, sessions are available starting from 1/1/2014. Thus, I used  4 classifiers: (1) XGB trained on all data,  (2) RandomForests trained on all data,  (3) XGB classifier trained on recent (aka fresh) data only, and (4) RandomForests trained on fresh data only. Classifiers' results formed the final prediction via weighted voting, and classifiers (3) and (4) had a higher weight.</p>\n\n<p>(2) I used one-hot encoding for the categorical features in the basic dataset, and simple frequency counts for the sessions data. To avoid overfitting, and speed up computations, I replaced all rare values , e.g. (device_type==Linux Desktop), with &quot;other&quot;.</p>\n\n<p>(3) To tune classifiers and evaluate the prediction I used CV over a simplified NDCG score function.</p>\n\n<p>You can find my code here:\n<a href=\"https://github.com/blerim153/kaggle_airbnb_new_user_bookings\">https://github.com/blerim153/kaggle_airbnb_new_user_bookings</a></p>",
      "rawMarkdown": "Hello Everyone,\r\n\r\nThis was a great competition ! \r\n\r\nThis is a description of my technique:\r\n(1)   \"Fresh\" data got a higher weight.\r\n An important characteristic of this competition is that there is a time cut-off between training and testing data -- training data end on 7/1/2014, and testing data start on that day. Moreover, sessions are available starting from 1/1/2014. Thus, I used  4 classifiers: (1) XGB trained on all data,  (2) RandomForests trained on all data,  (3) XGB classifier trained on recent (aka fresh) data only, and (4) RandomForests trained on fresh data only. Classifiers' results formed the final prediction via weighted voting, and classifiers (3) and (4) had a higher weight.\r\n\r\n(2) I used one-hot encoding for the categorical features in the basic dataset, and simple frequency counts for the sessions data. To avoid overfitting, and speed up computations, I replaced all rare values , e.g. (device_type==Linux Desktop), with \"other\".\r\n\r\n(3) To tune classifiers and evaluate the prediction I used CV over a simplified NDCG score function.\r\n\r\nYou can find my code here:\r\nhttps://github.com/blerim153/kaggle_airbnb_new_user_bookings",
      "votes": 1
    }
  ],
  "complete": true,
  "source": "live",
  "fetched_at": "2026-10-19T13:52:25Z",
  "competition_id": 4651,
  "competition_slug": "airbnb-recruiting-new-user-bookings",
  "team_id": 262645,
  "solution_url": "/competitions/airbnb-recruiting-new-user-bookings/writeups/blerim-code-sharing-60th-position-on-private-leade"
}