{
  "id": 2848,
  "title": "The home straight: go bagging",
  "url": "/competitions/predict-closed-questions-on-stack-overflow/discussion/2848",
  "author_name": "",
  "post_date": "2012-10-09T14:14:12.127Z",
  "votes": 1,
  "comment_count": 3,
  "views": 2855,
  "content": "<p>If you happen to have a few relatively good submission files lying around and haven't used ensembling so far, you might want to try averaging your solutions. Attached is a script for this. You will need to edit the script to set the names for input files\r\n and an output file.</p>",
  "messages": [
    {
      "id": "15347",
      "postDate": "10/09/2012 14:14:12",
      "content": "<p>If you happen to have a few relatively good submission files lying around and haven't used ensembling so far, you might want to try averaging your solutions. Attached is a script for this. You will need to edit the script to set the names for input files\r\n and an output file.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "15370",
      "postDate": "10/09/2012 23:58:34",
      "content": "<p>Neat.</p>\r\n<p>I didn't realise that bagging extremely similar models still works! (especially since my models are bags of random forests... which are of course bags of random trees...).</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "15371",
      "postDate": "10/10/2012 00:03:14",
      "content": "<p>I tried bagging with vw it looked great on the test set: 0.099 log loss. But on validation it was 0.19 - worse than unbagged models.</p>\r\n<p>I'm sad that I didn't try with vw before. I tried using it in EMC competition but I coudn't compile it, because of missing headers.</p>\r\n<p>Oh well It's 2 AM school day tomorrow. It was a nice competition.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "15435",
      "postDate": "10/11/2012 17:05:26",
      "content": "<p>bagging or no method will work in cases where the models you are trying to blend are similar!</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 15370,
      "author_name": "james30",
      "author_url": "",
      "post_date": "10/09/2012 23:58:34",
      "content": "<p>Neat.</p>\r\n<p>I didn't realise that bagging extremely similar models still works! (especially since my models are bags of random forests... which are of course bags of random trees...).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 15371,
      "author_name": "marbur",
      "author_url": "",
      "post_date": "10/10/2012 00:03:14",
      "content": "<p>I tried bagging with vw it looked great on the test set: 0.099 log loss. But on validation it was 0.19 - worse than unbagged models.</p>\r\n<p>I'm sad that I didn't try with vw before. I tried using it in EMC competition but I coudn't compile it, because of missing headers.</p>\r\n<p>Oh well It's 2 AM school day tomorrow. It was a nice competition.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 15435,
      "author_name": "rkirana",
      "author_url": "",
      "post_date": "10/11/2012 17:05:26",
      "content": "<p>bagging or no method will work in cases where the models you are trying to blend are similar!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "15347": "",
    "15370": "",
    "15371": "",
    "15435": ""
  },
  "source": "meta"
}