{
  "id": 159492,
  "title": "testset distribution from submission scores.",
  "url": "/competitions/birdsong-recognition/discussion/159492",
  "author_name": "Dhananjay Raut",
  "post_date": "2020-06-17T16:02:28.299000",
  "votes": 29,
  "comment_count": 2,
  "views": 0,
  "content": "<p>we know predicting <code>nocall</code> has score 0.54 which is avg of row wise micro F1.\n| row type|fraction|Precision|Recall| F1\n| ---|--- | --- |---|---|\nbird | a |0 | 0 | 0\nno-bird | 1-a  | 1 | 1 | 1</p>\n\n<p><code>\nfinal F1 = a * (F1bird)  + (1-a) (F1nobird) = 1 - a\n1 - a = 0.54\na = 0.46\n</code></p>\n\n<p><strong>46% test clips have birds singing.</strong></p>\n\n<p><strong>54% don't. (unlike train set where each clip contains at least one bird)</strong></p>",
  "messages": [
    {
      "id": 890632,
      "postDate": "2020-06-17T16:02:28.300Z",
      "content": "<p>we know predicting <code>nocall</code> has score 0.54 which is avg of row wise micro F1.\n| row type|fraction|Precision|Recall| F1\n| ---|--- | --- |---|---|\nbird | a |0 | 0 | 0\nno-bird | 1-a  | 1 | 1 | 1</p>\n\n<p><code>\nfinal F1 = a * (F1bird)  + (1-a) (F1nobird) = 1 - a\n1 - a = 0.54\na = 0.46\n</code></p>\n\n<p><strong>46% test clips have birds singing.</strong></p>\n\n<p><strong>54% don't. (unlike train set where each clip contains at least one bird)</strong></p>",
      "rawMarkdown": "we know predicting `nocall` has score 0.54 which is avg of row wise micro F1.\n| row type|fraction|Precision|Recall| F1\n| ---|--- | --- |---|---|\nbird | a |0 | 0 | 0\nno-bird | 1-a  | 1 | 1 | 1\n\n```\nfinal F1 = a * (F1bird)  + (1-a) (F1nobird) = 1 - a\n1 - a = 0.54\na = 0.46\n```\n\n**46% test clips have birds singing.**\n\n**54% don't. (unlike train set where each clip contains at least one bird)**",
      "votes": 29
    },
    {
      "id": 892852,
      "postDate": "2020-06-19T07:51:02.087Z",
      "content": "<p>That would be for the public leaderboard 27% of the data for test clips, not sure would be the same for private test. </p>",
      "rawMarkdown": "That would be for the public leaderboard 27% of the data for test clips, not sure would be the same for private test. ",
      "votes": 5
    },
    {
      "id": 892123,
      "postDate": "2020-06-18T17:10:09.587Z",
      "rawMarkdown": "",
      "votes": -2,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 892852,
      "author_name": "something4kag",
      "author_url": "",
      "post_date": "2020-06-19T07:51:02.087000",
      "content": "<p>That would be for the public leaderboard 27% of the data for test clips, not sure would be the same for private test. </p>",
      "votes": 5,
      "replies": []
    },
    {
      "id": 892123,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-06-18T17:10:09.587000",
      "content": "",
      "votes": -2,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "890632": "we know predicting `nocall` has score 0.54 which is avg of row wise micro F1.\n| row type|fraction|Precision|Recall| F1\n| ---|--- | --- |---|---|\nbird | a |0 | 0 | 0\nno-bird | 1-a  | 1 | 1 | 1\n\n```\nfinal F1 = a * (F1bird)  + (1-a) (F1nobird) = 1 - a\n1 - a = 0.54\na = 0.46\n```\n\n**46% test clips have birds singing.**\n\n**54% don't. (unlike train set where each clip contains at least one bird)**",
    "892852": "That would be for the public leaderboard 27% of the data for test clips, not sure would be the same for private test. ",
    "892123": ""
  }
}