{
  "id": 164936,
  "title": "Confused about Predicting Confidence Measure",
  "url": "/competitions/osic-pulmonary-fibrosis-progression/discussion/164936",
  "author_name": "",
  "post_date": "2020-07-08T00:04:12.584105700Z",
  "votes": 9,
  "comment_count": 3,
  "views": 0,
  "content": "<p>This might be a stupid question, but in the Evaluation section it states: </p>\n\n<blockquote>\n  <p>For each true FVC measurement, you will predict both an FVC and a confidence measure (standard deviation σ). </p>\n</blockquote>\n\n<p>Is \"standard deviation σ\" meant as just the symbol we should treat this prediction, or does it mean that the confidence measure is literally the standard deviation of the FVC? Thanks!</p>",
  "messages": [
    {
      "id": "919517",
      "postDate": "07/08/2020 00:04:12",
      "content": "<p>This might be a stupid question, but in the Evaluation section it states: </p>\n\n<blockquote>\n  <p>For each true FVC measurement, you will predict both an FVC and a confidence measure (standard deviation σ). </p>\n</blockquote>\n\n<p>Is \"standard deviation σ\" meant as just the symbol we should treat this prediction, or does it mean that the confidence measure is literally the standard deviation of the FVC? Thanks!</p>",
      "rawMarkdown": "This might be a stupid question, but in the Evaluation section it states: \n\n&gt; For each true FVC measurement, you will predict both an FVC and a confidence measure (standard deviation σ). \n\nIs \"standard deviation σ\" meant as just the symbol we should treat this prediction, or does it mean that the confidence measure is literally the standard deviation of the FVC? Thanks!",
      "votes": null
    },
    {
      "id": "919706",
      "postDate": "07/08/2020 04:27:37",
      "content": "<p>It is literally the standard deviation of FVC ( considering the last three values) , and then it is clipped to a max value of 70 </p>",
      "rawMarkdown": "It is literally the standard deviation of FVC ( considering the last three values) , and then it is clipped to a max value of 70",
      "votes": null
    },
    {
      "id": "932298",
      "postDate": "07/16/2020 23:15:27",
      "content": "<p>What do you mean about clipped the max value of 70? It is to choose the max value between the std dev and 70?</p>",
      "rawMarkdown": "What do you mean about clipped the max value of 70? It is to choose the max value between the std dev and 70?",
      "votes": null
    },
    {
      "id": "932347",
      "postDate": "07/17/2020 01:45:24",
      "content": "<p><a href=\"/rohanrao\">@rohanrao</a> answered my question in his notebook: <a href=\"https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood\">https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood</a>. </p>\n\n<p>The main point being that you <em>can</em> calculate the Confidence as the standard deviation, but this will result in poorer performance compared to using some smarter technique to predict the Confidence value.</p>\n\n<p>If you look at the graphs that <a href=\"/rohanrao\">@rohanrao</a> presents, you find that the Laplace Log Likelihood will exponentially punish you if you are extremely confident (lower confidence). Perhaps for this reason the competition clips the Confidence at 70.</p>",
      "rawMarkdown": "rohanrao answered my question in his notebook: https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood. \n\nThe main point being that you *can* calculate the Confidence as the standard deviation, but this will result in poorer performance compared to using some smarter technique to predict the Confidence value.\n\nIf you look at the graphs that @rohanrao presents, you find that the Laplace Log Likelihood will exponentially punish you if you are extremely confident (lower confidence). Perhaps for this reason the competition clips the Confidence at 70.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 919706,
      "author_name": "venkat555",
      "author_url": "",
      "post_date": "07/08/2020 04:27:37",
      "content": "<p>It is literally the standard deviation of FVC ( considering the last three values) , and then it is clipped to a max value of 70 </p>",
      "votes": null,
      "replies": [
        {
          "id": 932298,
          "author_name": "natalia93",
          "author_url": "",
          "post_date": "07/16/2020 23:15:27",
          "content": "<p>What do you mean about clipped the max value of 70? It is to choose the max value between the std dev and 70?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 932347,
          "author_name": "jjinho",
          "author_url": "",
          "post_date": "07/17/2020 01:45:24",
          "content": "<p><a href=\"/rohanrao\">@rohanrao</a> answered my question in his notebook: <a href=\"https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood\">https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood</a>. </p>\n\n<p>The main point being that you <em>can</em> calculate the Confidence as the standard deviation, but this will result in poorer performance compared to using some smarter technique to predict the Confidence value.</p>\n\n<p>If you look at the graphs that <a href=\"/rohanrao\">@rohanrao</a> presents, you find that the Laplace Log Likelihood will exponentially punish you if you are extremely confident (lower confidence). Perhaps for this reason the competition clips the Confidence at 70.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "919517": "This might be a stupid question, but in the Evaluation section it states: \n\n&gt; For each true FVC measurement, you will predict both an FVC and a confidence measure (standard deviation σ). \n\nIs \"standard deviation σ\" meant as just the symbol we should treat this prediction, or does it mean that the confidence measure is literally the standard deviation of the FVC? Thanks!",
    "919706": "It is literally the standard deviation of FVC ( considering the last three values) , and then it is clipped to a max value of 70",
    "932298": "What do you mean about clipped the max value of 70? It is to choose the max value between the std dev and 70?",
    "932347": "rohanrao answered my question in his notebook: https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood. \n\nThe main point being that you *can* calculate the Confidence as the standard deviation, but this will result in poorer performance compared to using some smarter technique to predict the Confidence value.\n\nIf you look at the graphs that @rohanrao presents, you find that the Laplace Log Likelihood will exponentially punish you if you are extremely confident (lower confidence). Perhaps for this reason the competition clips the Confidence at 70."
  },
  "source": "meta"
}