{
  "id": 70053,
  "title": "kaggle F1 score question",
  "url": "/competitions/human-protein-atlas-image-classification/discussion/70053",
  "author_name": "",
  "post_date": "2018-10-30T09:51:46.489656700Z",
  "votes": 10,
  "comment_count": 3,
  "views": 0,
  "content": "<p>which is correct?</p>\n\n<p>[1]  F1 computed once over all images</p>\n\n<p>[2] F1 computed for each image. Then all F1's are average.</p>\n\n<p>I recall that previous f2 score of kaggle amazon forest challenge uses method [2] \n<a href=\"https://www.kaggle.com/c/planet-understanding-the-amazon-from-space#evaluation\">https://www.kaggle.com/c/planet-understanding-the-amazon-from-space#evaluation</a></p>",
  "messages": [
    {
      "id": "412493",
      "postDate": "10/30/2018 09:51:46",
      "content": "<p>which is correct?</p>\n\n<p>[1]  F1 computed once over all images</p>\n\n<p>[2] F1 computed for each image. Then all F1's are average.</p>\n\n<p>I recall that previous f2 score of kaggle amazon forest challenge uses method [2] \n<a href=\"https://www.kaggle.com/c/planet-understanding-the-amazon-from-space#evaluation\">https://www.kaggle.com/c/planet-understanding-the-amazon-from-space#evaluation</a></p>",
      "rawMarkdown": "which is correct?\n\n[1]  F1 computed once over all images\n\n[2] F1 computed for each image. Then all F1's are average.\n\nI recall that previous f2 score of kaggle amazon forest challenge uses method [2] \nhttps://www.kaggle.com/c/planet-understanding-the-amazon-from-space#evaluation",
      "votes": null
    },
    {
      "id": "412507",
      "postDate": "10/30/2018 10:19:15",
      "content": "<p>I think the 'macro' version we are using is [3] F1 computed separately for each class over all images, then averaged</p>",
      "rawMarkdown": "I think the 'macro' version we are using is [3] F1 computed separately for each class over all images, then averaged",
      "votes": null
    },
    {
      "id": "412511",
      "postDate": "10/30/2018 10:23:22",
      "content": "<p>@Russ W</p>\n\n<p>Thanks for the answer!</p>",
      "rawMarkdown": "Russ W\n\nThanks for the answer!",
      "votes": null
    },
    {
      "id": "412516",
      "postDate": "10/30/2018 10:28:50",
      "content": "<p>Hello, Heng CherKeng. Kaggle have a two versions of F-Score for multilabel problems: <strong>Mean F-Score</strong> and <strong>Macro F-Score</strong>. In this competition we got <strong>Macro F-Score</strong> that is the same as <code>sklearn.metrics.f1_score</code> with <code>average=\"macro\"</code> parameter. And it calculates metrics for each label, and then finds their unweighted mean. By the way, <strong>Mean F-Score</strong> is the same as <code>sklearn.metrics.f1_score</code> with <code>average=\"samples\"</code> parameter. And this one computes metrics for each instance, and then finds their average.</p>",
      "rawMarkdown": "Hello, Heng CherKeng. Kaggle have a two versions of F-Score for multilabel problems: **Mean F-Score** and **Macro F-Score**. In this competition we got **Macro F-Score** that is the same as `sklearn.metrics.f1_score` with `average=\"macro\"` parameter. And it calculates metrics for each label, and then finds their unweighted mean. By the way, **Mean F-Score** is the same as `sklearn.metrics.f1_score` with `average=\"samples\"` parameter. And this one computes metrics for each instance, and then finds their average.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 412507,
      "author_name": "sasrdw",
      "author_url": "",
      "post_date": "10/30/2018 10:19:15",
      "content": "<p>I think the 'macro' version we are using is [3] F1 computed separately for each class over all images, then averaged</p>",
      "votes": null,
      "replies": [
        {
          "id": 412511,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "10/30/2018 10:23:22",
          "content": "<p>@Russ W</p>\n\n<p>Thanks for the answer!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 412516,
      "author_name": "sggpls",
      "author_url": "",
      "post_date": "10/30/2018 10:28:50",
      "content": "<p>Hello, Heng CherKeng. Kaggle have a two versions of F-Score for multilabel problems: <strong>Mean F-Score</strong> and <strong>Macro F-Score</strong>. In this competition we got <strong>Macro F-Score</strong> that is the same as <code>sklearn.metrics.f1_score</code> with <code>average=\"macro\"</code> parameter. And it calculates metrics for each label, and then finds their unweighted mean. By the way, <strong>Mean F-Score</strong> is the same as <code>sklearn.metrics.f1_score</code> with <code>average=\"samples\"</code> parameter. And this one computes metrics for each instance, and then finds their average.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "412493": "which is correct?\n\n[1]  F1 computed once over all images\n\n[2] F1 computed for each image. Then all F1's are average.\n\nI recall that previous f2 score of kaggle amazon forest challenge uses method [2] \nhttps://www.kaggle.com/c/planet-understanding-the-amazon-from-space#evaluation",
    "412507": "I think the 'macro' version we are using is [3] F1 computed separately for each class over all images, then averaged",
    "412511": "Russ W\n\nThanks for the answer!",
    "412516": "Hello, Heng CherKeng. Kaggle have a two versions of F-Score for multilabel problems: **Mean F-Score** and **Macro F-Score**. In this competition we got **Macro F-Score** that is the same as `sklearn.metrics.f1_score` with `average=\"macro\"` parameter. And it calculates metrics for each label, and then finds their unweighted mean. By the way, **Mean F-Score** is the same as `sklearn.metrics.f1_score` with `average=\"samples\"` parameter. And this one computes metrics for each instance, and then finds their average."
  },
  "source": "meta"
}