{
  "id": 21503,
  "title": "What score should my xgboost get to beat the script?",
  "url": "/competitions/expedia-hotel-recommendations/discussion/21503",
  "author_name": "",
  "post_date": "2016-06-08T02:08:05.110Z",
  "votes": null,
  "comment_count": 1,
  "views": 402,
  "content": "<p>Now my xgboost score is LB 0.271 which takes me a lot of time to train. I'm wandering what xbgboost score combined with leakage information could help me beat the script.</p>",
  "messages": [
    {
      "id": "122883",
      "postDate": "06/08/2016 02:08:05",
      "content": "<p>Now my xgboost score is LB 0.271 which takes me a lot of time to train. I'm wandering what xbgboost score combined with leakage information could help me beat the script.</p>",
      "rawMarkdown": "Now my xgboost score is LB 0.271 which takes me a lot of time to train. I'm wandering what xbgboost score combined with leakage information could help me beat the script.",
      "votes": null
    },
    {
      "id": "122895",
      "postDate": "06/08/2016 04:28:59",
      "content": "<p>FengLi,</p>\n\n<p>Assuming you predict leakage and non-leakage set separately, your LB would be average of scores of the leakage and non-leakage set, weighted by size of each set</p>\n\n<p>with below assumptions:\n- you use XGB to predict non-leakage data\n- leakage set is 1/3 of test data with score of 0.9\n- I don't know what script you try to beat but let's say you want 0.5 on LB</p>\n\n<p>then you need about 0.3 for your XGB predictions on non-leakage data</p>",
      "rawMarkdown": "FengLi,\r\n\r\nAssuming you predict leakage and non-leakage set separately, your LB would be average of scores of the leakage and non-leakage set, weighted by size of each set\r\n\r\n with below assumptions:\r\n- you use XGB to predict non-leakage data\r\n- leakage set is 1/3 of test data with score of 0.9\r\n- I don't know what script you try to beat but let's say you want 0.5 on LB\r\n\r\nthen you need about 0.3 for your XGB predictions on non-leakage data",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 122895,
      "author_name": "kenzheng",
      "author_url": "",
      "post_date": "06/08/2016 04:28:59",
      "content": "<p>FengLi,</p>\n\n<p>Assuming you predict leakage and non-leakage set separately, your LB would be average of scores of the leakage and non-leakage set, weighted by size of each set</p>\n\n<p>with below assumptions:\n- you use XGB to predict non-leakage data\n- leakage set is 1/3 of test data with score of 0.9\n- I don't know what script you try to beat but let's say you want 0.5 on LB</p>\n\n<p>then you need about 0.3 for your XGB predictions on non-leakage data</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "122883": "Now my xgboost score is LB 0.271 which takes me a lot of time to train. I'm wandering what xbgboost score combined with leakage information could help me beat the script.",
    "122895": "FengLi,\r\n\r\nAssuming you predict leakage and non-leakage set separately, your LB would be average of scores of the leakage and non-leakage set, weighted by size of each set\r\n\r\n with below assumptions:\r\n- you use XGB to predict non-leakage data\r\n- leakage set is 1/3 of test data with score of 0.9\r\n- I don't know what script you try to beat but let's say you want 0.5 on LB\r\n\r\nthen you need about 0.3 for your XGB predictions on non-leakage data"
  },
  "source": "meta"
}