{
  "id": 1996,
  "title": "cross validation",
  "url": "/competitions/FacebookRecruiting/discussion/1996",
  "author_name": "",
  "post_date": "2012-06-06T02:27:29.280Z",
  "votes": null,
  "comment_count": 3,
  "views": 3388,
  "content": "<p>I have no idea on how to conduct cross validation for this task.</p>\r\n<p>Deleting more edges?</p>\r\n<p>Anyone who gives some hint? Thanks.</p>",
  "messages": [
    {
      "id": "11415",
      "postDate": "06/06/2012 02:27:29",
      "content": "<p>I have no idea on how to conduct cross validation for this task.</p>\r\n<p>Deleting more edges?</p>\r\n<p>Anyone who gives some hint? Thanks.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "11448",
      "postDate": "06/07/2012 19:20:12",
      "content": "<p>My guess is to use a subset of the training data to build your model and then use the model to predict on the other nodes (the ones you left out in the model-building step). Then, since you have the true edges for these cases, you can see how well your predictions\r\n compare to reality.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "11455",
      "postDate": "06/08/2012 03:30:57",
      "content": "<p>It's an interesting question. The training set has already been trimmed, so if you trim it further and then run some algorithm, it may predict edges from either of the two trimmings, but you'll only know if they were from your trimming.</p>\r\n<p>I guess it would be ideal to experiment with a complete graph with the same properties as the complete graph for this competition, but I don't know where to get such a graph to use.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "11471",
      "postDate": "06/08/2012 14:11:40",
      "content": "<p>[quote=Aaron Schumacher;11455]</p>\r\n<p>It's an interesting question. The training set has already been trimmed, so if you trim it further and then run some algorithm, it may predict edges from either of the two trimmings, but you'll only know if they were from your trimming.</p>\r\n<p>I guess it would be ideal to experiment with a complete graph with the same properties as the complete graph for this competition, but I don't know where to get such a graph to use.</p>\r\n<p>[/quote]</p>\r\n<p>agree.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 11448,
      "author_name": "randomforestfanatic",
      "author_url": "",
      "post_date": "06/07/2012 19:20:12",
      "content": "<p>My guess is to use a subset of the training data to build your model and then use the model to predict on the other nodes (the ones you left out in the model-building step). Then, since you have the true edges for these cases, you can see how well your predictions\r\n compare to reality.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 11455,
      "author_name": "ajschumacher",
      "author_url": "",
      "post_date": "06/08/2012 03:30:57",
      "content": "<p>It's an interesting question. The training set has already been trimmed, so if you trim it further and then run some algorithm, it may predict edges from either of the two trimmings, but you'll only know if they were from your trimming.</p>\r\n<p>I guess it would be ideal to experiment with a complete graph with the same properties as the complete graph for this competition, but I don't know where to get such a graph to use.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 11471,
      "author_name": "waynezhang",
      "author_url": "",
      "post_date": "06/08/2012 14:11:40",
      "content": "<p>[quote=Aaron Schumacher;11455]</p>\r\n<p>It's an interesting question. The training set has already been trimmed, so if you trim it further and then run some algorithm, it may predict edges from either of the two trimmings, but you'll only know if they were from your trimming.</p>\r\n<p>I guess it would be ideal to experiment with a complete graph with the same properties as the complete graph for this competition, but I don't know where to get such a graph to use.</p>\r\n<p>[/quote]</p>\r\n<p>agree.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "11415": "",
    "11448": "",
    "11455": "",
    "11471": ""
  },
  "source": "meta"
}