{
  "id": 35879,
  "title": "Stage 2 dataset - new Threat objects as well as new people?",
  "url": "/competitions/passenger-screening-algorithm-challenge/discussion/35879",
  "author_name": "",
  "post_date": "2017-07-06T15:22:39.941311800Z",
  "votes": 4,
  "comment_count": 2,
  "views": 2,
  "content": "<p><strong>A question for the competition admin individuals</strong></p>\n\n<p>In the stage 2 dataset we know that different people  will be used - to wit this paragraph in the data description:</p>\n\n<p><em>The volunteers used in the first and second stage of the competition will be different (i.e. your algorithm should generalize to unseen people). In addition, you should not make assumptions about the number, distribution, or location of threats in the second stage.</em></p>\n\n<p>But how about <strong>different threat objects</strong> ? Implicitly, reading the above suggests that the threat objects will be the same in stage 1 and stage 2.</p>\n\n<p>Am I right ?</p>",
  "messages": [
    {
      "id": "199903",
      "postDate": "07/06/2017 15:22:39",
      "content": "<p><strong>A question for the competition admin individuals</strong></p>\n\n<p>In the stage 2 dataset we know that different people  will be used - to wit this paragraph in the data description:</p>\n\n<p><em>The volunteers used in the first and second stage of the competition will be different (i.e. your algorithm should generalize to unseen people). In addition, you should not make assumptions about the number, distribution, or location of threats in the second stage.</em></p>\n\n<p>But how about <strong>different threat objects</strong> ? Implicitly, reading the above suggests that the threat objects will be the same in stage 1 and stage 2.</p>\n\n<p>Am I right ?</p>",
      "rawMarkdown": "**A question for the competition admin individuals**\n\nIn the stage 2 dataset we know that different people  will be used - to wit this paragraph in the data description:\n\n*The volunteers used in the first and second stage of the competition will be different (i.e. your algorithm should generalize to unseen people). In addition, you should not make assumptions about the number, distribution, or location of threats in the second stage.*\n\nBut how about **different threat objects** ? Implicitly, reading the above suggests that the threat objects will be the same in stage 1 and stage 2.\n\nAm I right ?",
      "votes": null
    },
    {
      "id": "221360",
      "postDate": "09/15/2017 00:46:42",
      "content": "<p>This may actually be the most important question in this competition.</p>\n\n<p>Suppose that <code>x</code> is the fraction of new threat objects in stage 2.</p>\n\n<p>If <code>x = 0</code>, then models that assume that <code>x = 1</code> (that is, those that try to generalize to previously unseen threats) will be disadvantaged.</p>\n\n<p>If <code>x = 1</code>, then models that assume that <code>x = 0</code> will be disadvantaged.</p>\n\n<p>A big part of the competition, to my mind, is reduced to guessing <code>x</code>. It might be best to include <code>x</code> in the official task description instead.</p>",
      "rawMarkdown": "This may actually be the most important question in this competition.\n\nSuppose that `x` is the fraction of new threat objects in stage 2.\n\nIf `x = 0`, then models that assume that `x = 1` (that is, those that try to generalize to previously unseen threats) will be disadvantaged.\n\nIf `x = 1`, then models that assume that `x = 0` will be disadvantaged.\n\nA big part of the competition, to my mind, is reduced to guessing `x`. It might be best to include `x` in the official task description instead.",
      "votes": null
    },
    {
      "id": "221507",
      "postDate": "09/15/2017 14:11:26",
      "content": "<p>The threat objects for both stages were chosen using the same methodology and will therefore be similar.</p>",
      "rawMarkdown": "The threat objects for both stages were chosen using the same methodology and will therefore be similar.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 221360,
      "author_name": "olegtrott",
      "author_url": "",
      "post_date": "09/15/2017 00:46:42",
      "content": "<p>This may actually be the most important question in this competition.</p>\n\n<p>Suppose that <code>x</code> is the fraction of new threat objects in stage 2.</p>\n\n<p>If <code>x = 0</code>, then models that assume that <code>x = 1</code> (that is, those that try to generalize to previously unseen threats) will be disadvantaged.</p>\n\n<p>If <code>x = 1</code>, then models that assume that <code>x = 0</code> will be disadvantaged.</p>\n\n<p>A big part of the competition, to my mind, is reduced to guessing <code>x</code>. It might be best to include <code>x</code> in the official task description instead.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 221507,
      "author_name": "wcukierski",
      "author_url": "",
      "post_date": "09/15/2017 14:11:26",
      "content": "<p>The threat objects for both stages were chosen using the same methodology and will therefore be similar.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "199903": "**A question for the competition admin individuals**\n\nIn the stage 2 dataset we know that different people  will be used - to wit this paragraph in the data description:\n\n*The volunteers used in the first and second stage of the competition will be different (i.e. your algorithm should generalize to unseen people). In addition, you should not make assumptions about the number, distribution, or location of threats in the second stage.*\n\nBut how about **different threat objects** ? Implicitly, reading the above suggests that the threat objects will be the same in stage 1 and stage 2.\n\nAm I right ?",
    "221360": "This may actually be the most important question in this competition.\n\nSuppose that `x` is the fraction of new threat objects in stage 2.\n\nIf `x = 0`, then models that assume that `x = 1` (that is, those that try to generalize to previously unseen threats) will be disadvantaged.\n\nIf `x = 1`, then models that assume that `x = 0` will be disadvantaged.\n\nA big part of the competition, to my mind, is reduced to guessing `x`. It might be best to include `x` in the official task description instead.",
    "221507": "The threat objects for both stages were chosen using the same methodology and will therefore be similar."
  },
  "source": "meta"
}