{
  "id": 48486,
  "title": "Evaluation bins",
  "url": "/competitions/data-science-bowl-2018/discussion/48486",
  "author_name": "",
  "post_date": "2018-01-28T14:12:05.775632100Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi, I’m trying to understand the evaluation formula. What is the effect of the threshold bins (0.05 intervals for IOU) ? How do they impact the final score?</p>",
  "messages": [
    {
      "id": "275220",
      "postDate": "01/28/2018 14:12:05",
      "content": "<p>Hi, I’m trying to understand the evaluation formula. What is the effect of the threshold bins (0.05 intervals for IOU) ? How do they impact the final score?</p>",
      "rawMarkdown": "Hi, I’m trying to understand the evaluation formula. What is the effect of the threshold bins (0.05 intervals for IOU) ? How do they impact the final score?",
      "votes": null
    },
    {
      "id": "275380",
      "postDate": "01/29/2018 01:44:33",
      "content": "<p>The printout in <a href=\"https://www.kaggle.com/wcukierski/example-metric-implementation\">William Cukierski's implementation of the metric</a> is helpful to understand this.  Essentially, at higher thresholds true positives are almost never found, only false positives and false negatives.  This is because the IOU threshold for matching objects is too high, or too strict.  At lower thresholds, true positives are almost always found, and very few false positives and false negatives.  This is because the IOU threshold for matching objects is too low, or too permissive.  Rather than pick a particular IOU threshold, which could work well for some images but not others, an overage over IOU thresholds is used.</p>",
      "rawMarkdown": "The printout in [William Cukierski's implementation of the metric][1] is helpful to understand this.  Essentially, at higher thresholds true positives are almost never found, only false positives and false negatives.  This is because the IOU threshold for matching objects is too high, or too strict.  At lower thresholds, true positives are almost always found, and very few false positives and false negatives.  This is because the IOU threshold for matching objects is too low, or too permissive.  Rather than pick a particular IOU threshold, which could work well for some images but not others, an overage over IOU thresholds is used.\n\n\n  [1]: https://www.kaggle.com/wcukierski/example-metric-implementation",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 275380,
      "author_name": "nathanrm",
      "author_url": "",
      "post_date": "01/29/2018 01:44:33",
      "content": "<p>The printout in <a href=\"https://www.kaggle.com/wcukierski/example-metric-implementation\">William Cukierski's implementation of the metric</a> is helpful to understand this.  Essentially, at higher thresholds true positives are almost never found, only false positives and false negatives.  This is because the IOU threshold for matching objects is too high, or too strict.  At lower thresholds, true positives are almost always found, and very few false positives and false negatives.  This is because the IOU threshold for matching objects is too low, or too permissive.  Rather than pick a particular IOU threshold, which could work well for some images but not others, an overage over IOU thresholds is used.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "275220": "Hi, I’m trying to understand the evaluation formula. What is the effect of the threshold bins (0.05 intervals for IOU) ? How do they impact the final score?",
    "275380": "The printout in [William Cukierski's implementation of the metric][1] is helpful to understand this.  Essentially, at higher thresholds true positives are almost never found, only false positives and false negatives.  This is because the IOU threshold for matching objects is too high, or too strict.  At lower thresholds, true positives are almost always found, and very few false positives and false negatives.  This is because the IOU threshold for matching objects is too low, or too permissive.  Rather than pick a particular IOU threshold, which could work well for some images but not others, an overage over IOU thresholds is used.\n\n\n  [1]: https://www.kaggle.com/wcukierski/example-metric-implementation"
  },
  "source": "meta"
}