{
  "id": 68328,
  "title": "Clear definition of F2 metric ",
  "url": "/competitions/inclusive-images-challenge/discussion/68328",
  "author_name": "",
  "post_date": "2018-10-11T14:20:40.681836200Z",
  "votes": 3,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Can anyone from organization team formalise of F2 metric for multi-label task? It is not clear if it is calculated by number of classes or number of samples. </p>",
  "messages": [
    {
      "id": "402322",
      "postDate": "10/11/2018 14:20:40",
      "content": "<p>Can anyone from organization team formalise of F2 metric for multi-label task? It is not clear if it is calculated by number of classes or number of samples. </p>",
      "rawMarkdown": "Can anyone from organization team formalise of F2 metric for multi-label task? It is not clear if it is calculated by number of classes or number of samples.",
      "votes": null
    },
    {
      "id": "402334",
      "postDate": "10/11/2018 14:39:14",
      "content": "<p>An F2 score is calculated for each sample based on the precision, recall of the predictions for that sample. \nThe scores are then averaged over samples.</p>\n\n<p>For example:\nIf sample 1 has labels {A, B} and you predict {A, B, C}, you have an F2 score of 10/11 on sample 1.\nHopefully my math is correct there. Precision=2/3, Recall=1.0 -&gt; F2=(2^2+1)*(2/3)*1 / (2^2 * (2/3) + 1) </p>\n\n<p>If sample 2 has labels {C} and you predict {C}, you have an F2 score of 1.0 on sample 2.</p>\n\n<p>If the test set consisted of just those two samples, then the F2 score for the test set would be the average over the sample: i.e. (1.0 + 10/11) / 2 = 21/11</p>",
      "rawMarkdown": "An F2 score is calculated for each sample based on the precision, recall of the predictions for that sample. \nThe scores are then averaged over samples.\n\nFor example:\nIf sample 1 has labels {A, B} and you predict {A, B, C}, you have an F2 score of 10/11 on sample 1.\nHopefully my math is correct there. Precision=2/3, Recall=1.0 -&gt; F2=(2^2+1)*(2/3)*1 / (2^2 * (2/3) + 1) \n\nIf sample 2 has labels {C} and you predict {C}, you have an F2 score of 1.0 on sample 2.\n\nIf the test set consisted of just those two samples, then the F2 score for the test set would be the average over the sample: i.e. (1.0 + 10/11) / 2 = 21/11",
      "votes": null
    },
    {
      "id": "402352",
      "postDate": "10/11/2018 14:56:37",
      "content": "<p>Thanks!</p>",
      "rawMarkdown": "Thanks!",
      "votes": null
    },
    {
      "id": "404549",
      "postDate": "10/16/2018 00:25:51",
      "content": "<p>Shouldn't it be 21/22? )))</p>",
      "rawMarkdown": "Shouldn't it be 21/22? )))",
      "votes": null
    },
    {
      "id": "404553",
      "postDate": "10/16/2018 00:43:58",
      "content": "<p>Good catch. Yes, that's right.</p>",
      "rawMarkdown": "Good catch. Yes, that's right.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 402334,
      "author_name": "yonihalpern",
      "author_url": "",
      "post_date": "10/11/2018 14:39:14",
      "content": "<p>An F2 score is calculated for each sample based on the precision, recall of the predictions for that sample. \nThe scores are then averaged over samples.</p>\n\n<p>For example:\nIf sample 1 has labels {A, B} and you predict {A, B, C}, you have an F2 score of 10/11 on sample 1.\nHopefully my math is correct there. Precision=2/3, Recall=1.0 -&gt; F2=(2^2+1)*(2/3)*1 / (2^2 * (2/3) + 1) </p>\n\n<p>If sample 2 has labels {C} and you predict {C}, you have an F2 score of 1.0 on sample 2.</p>\n\n<p>If the test set consisted of just those two samples, then the F2 score for the test set would be the average over the sample: i.e. (1.0 + 10/11) / 2 = 21/11</p>",
      "votes": null,
      "replies": [
        {
          "id": 402352,
          "author_name": "oktai15",
          "author_url": "",
          "post_date": "10/11/2018 14:56:37",
          "content": "<p>Thanks!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 404549,
          "author_name": "tetelias",
          "author_url": "",
          "post_date": "10/16/2018 00:25:51",
          "content": "<p>Shouldn't it be 21/22? )))</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 404553,
          "author_name": "yonihalpern",
          "author_url": "",
          "post_date": "10/16/2018 00:43:58",
          "content": "<p>Good catch. Yes, that's right.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "402322": "Can anyone from organization team formalise of F2 metric for multi-label task? It is not clear if it is calculated by number of classes or number of samples.",
    "402334": "An F2 score is calculated for each sample based on the precision, recall of the predictions for that sample. \nThe scores are then averaged over samples.\n\nFor example:\nIf sample 1 has labels {A, B} and you predict {A, B, C}, you have an F2 score of 10/11 on sample 1.\nHopefully my math is correct there. Precision=2/3, Recall=1.0 -&gt; F2=(2^2+1)*(2/3)*1 / (2^2 * (2/3) + 1) \n\nIf sample 2 has labels {C} and you predict {C}, you have an F2 score of 1.0 on sample 2.\n\nIf the test set consisted of just those two samples, then the F2 score for the test set would be the average over the sample: i.e. (1.0 + 10/11) / 2 = 21/11",
    "402352": "Thanks!",
    "404549": "Shouldn't it be 21/22? )))",
    "404553": "Good catch. Yes, that's right."
  },
  "source": "meta"
}