{
  "id": 48632,
  "title": "Question about the metric",
  "url": "/competitions/data-science-bowl-2018/discussion/48632",
  "author_name": "",
  "post_date": "2018-01-30T20:47:35.347284600Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I need clarification about how the threshold is incorporated into the computation of the average precision of a single image. The formula is presented as:</p>\n\n<p>&gt; \\frac{1}{|thresholds|} \\sum_t \\frac{TP(t)}{TP(t) + FP(t) + FN(t)}</p>\n\n<p>I don't understand three things about the first term: </p>\n\n<ul>\n<li><p>Why is it plural? Shouldn't it be \"threshold\"?</p></li>\n<li><p>Why is it outside the sum? Shouldn't it be inside?</p></li>\n<li><p>Why the absolute value bars? There is no need, thresholds are not negative.</p></li>\n</ul>\n\n<p>I think the formula should probably be:</p>\n\n<p>&gt; \\sum_t \\frac{1}{threshold} \\frac{TP(t)}{TP(t) + FP(t) + FN(t)}</p>\n\n<p>Somebody please confirm or correct.</p>",
  "messages": [
    {
      "id": "276153",
      "postDate": "01/30/2018 20:47:35",
      "content": "<p>I need clarification about how the threshold is incorporated into the computation of the average precision of a single image. The formula is presented as:</p>\n\n<p>&gt; \\frac{1}{|thresholds|} \\sum_t \\frac{TP(t)}{TP(t) + FP(t) + FN(t)}</p>\n\n<p>I don't understand three things about the first term: </p>\n\n<ul>\n<li><p>Why is it plural? Shouldn't it be \"threshold\"?</p></li>\n<li><p>Why is it outside the sum? Shouldn't it be inside?</p></li>\n<li><p>Why the absolute value bars? There is no need, thresholds are not negative.</p></li>\n</ul>\n\n<p>I think the formula should probably be:</p>\n\n<p>&gt; \\sum_t \\frac{1}{threshold} \\frac{TP(t)}{TP(t) + FP(t) + FN(t)}</p>\n\n<p>Somebody please confirm or correct.</p>",
      "rawMarkdown": "I need clarification about how the threshold is incorporated into the computation of the average precision of a single image. The formula is presented as:\n\n&gt; \\frac{1}{|thresholds|} \\sum_t \\frac{TP(t)}{TP(t) + FP(t) + FN(t)}\n\nI don't understand three things about the first term: \n\n * Why is it plural? Shouldn't it be \"threshold\"?\n\n * Why is it outside the sum? Shouldn't it be inside?\n\n * Why the absolute value bars? There is no need, thresholds are not negative.\n\nI think the formula should probably be:\n\n&gt; \\sum_t \\frac{1}{threshold} \\frac{TP(t)}{TP(t) + FP(t) + FN(t)}\n\nSomebody please confirm or correct.",
      "votes": null
    },
    {
      "id": "276171",
      "postDate": "01/30/2018 22:09:23",
      "content": "<p><em>Thresholds</em> would probably be more explicit as \"# of thresholds\". In this case, it comes out to 1/10. Computationally, it doesn't matter whether the 1/10 is inside the sum or outside. I agree the absolute value bars aren't needed, since there will always be more than one threshold.</p>\n\n<p>In other words, the evaluation metric for an image is the unweighted mean of the precision values at the different thresholds</p>",
      "rawMarkdown": "*Thresholds* would probably be more explicit as \"# of thresholds\". In this case, it comes out to 1/10. Computationally, it doesn't matter whether the 1/10 is inside the sum or outside. I agree the absolute value bars aren't needed, since there will always be more than one threshold.\n\nIn other words, the evaluation metric for an image is the unweighted mean of the precision values at the different thresholds",
      "votes": null
    },
    {
      "id": "276173",
      "postDate": "01/30/2018 22:11:41",
      "content": "<p>Thanks!</p>",
      "rawMarkdown": "Thanks!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 276171,
      "author_name": "bencopeland",
      "author_url": "",
      "post_date": "01/30/2018 22:09:23",
      "content": "<p><em>Thresholds</em> would probably be more explicit as \"# of thresholds\". In this case, it comes out to 1/10. Computationally, it doesn't matter whether the 1/10 is inside the sum or outside. I agree the absolute value bars aren't needed, since there will always be more than one threshold.</p>\n\n<p>In other words, the evaluation metric for an image is the unweighted mean of the precision values at the different thresholds</p>",
      "votes": null,
      "replies": [
        {
          "id": 276173,
          "author_name": "harald",
          "author_url": "",
          "post_date": "01/30/2018 22:11:41",
          "content": "<p>Thanks!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "276153": "I need clarification about how the threshold is incorporated into the computation of the average precision of a single image. The formula is presented as:\n\n&gt; \\frac{1}{|thresholds|} \\sum_t \\frac{TP(t)}{TP(t) + FP(t) + FN(t)}\n\nI don't understand three things about the first term: \n\n * Why is it plural? Shouldn't it be \"threshold\"?\n\n * Why is it outside the sum? Shouldn't it be inside?\n\n * Why the absolute value bars? There is no need, thresholds are not negative.\n\nI think the formula should probably be:\n\n&gt; \\sum_t \\frac{1}{threshold} \\frac{TP(t)}{TP(t) + FP(t) + FN(t)}\n\nSomebody please confirm or correct.",
    "276171": "*Thresholds* would probably be more explicit as \"# of thresholds\". In this case, it comes out to 1/10. Computationally, it doesn't matter whether the 1/10 is inside the sum or outside. I agree the absolute value bars aren't needed, since there will always be more than one threshold.\n\nIn other words, the evaluation metric for an image is the unweighted mean of the precision values at the different thresholds",
    "276173": "Thanks!"
  },
  "source": "meta"
}