{
  "id": 155933,
  "title": "Is the test data split stratified?",
  "url": "/competitions/global-wheat-detection/discussion/155933",
  "author_name": "",
  "post_date": "2020-06-03T15:40:43.044830100Z",
  "votes": 4,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I understand that the test data split is 70:30, 70% being for public LB, and 30% for private LB. Does anyone know if the split is stratified across source?</p>\n\n<p><strong>EDIT</strong> It was pointed out to me that the split is actually 62:38</p>",
  "messages": [
    {
      "id": "872898",
      "postDate": "06/03/2020 15:40:43",
      "content": "<p>I understand that the test data split is 70:30, 70% being for public LB, and 30% for private LB. Does anyone know if the split is stratified across source?</p>\n\n<p><strong>EDIT</strong> It was pointed out to me that the split is actually 62:38</p>",
      "rawMarkdown": "I understand that the test data split is 70:30, 70% being for public LB, and 30% for private LB. Does anyone know if the split is stratified across source?\n\n**EDIT** It was pointed out to me that the split is actually 62:38",
      "votes": null
    },
    {
      "id": "872939",
      "postDate": "06/03/2020 16:04:38",
      "content": "<p>From the leaderboard page:</p>\n\n<blockquote>\n  <p>This leaderboard is calculated with approximately 62% of the test data.\n  The final results will be based on the other 38%, so the final standings may be different.</p>\n</blockquote>\n\n<p>62% is a weird number for splitting data. Here is my calculation (the overall number of test samples are from the host's publication, everything else is just my opinion.)</p>\n\n<p>| Source | Overall | Public | Private  |\n|:-------:|:-------:|:-------:|:-------:|\n|  Utokyo1 | 994 | 794| 200  |\n|  Utokyo2 | 120 | 0 | 120  |\n|  NAU1 | 20 | 0| 20  |\n|  UQ1 | 142 | 0| 142  |</p>\n\n<p>Again, this is my guess, I have 0 proof!!</p>",
      "rawMarkdown": "From the leaderboard page:\n&gt; This leaderboard is calculated with approximately 62% of the test data.\n&gt; The final results will be based on the other 38%, so the final standings may be different.\n\n62% is a weird number for splitting data. Here is my calculation (the overall number of test samples are from the host's publication, everything else is just my opinion.)\n\n| Source | Overall | Public | Private  |\n|:-------:|:-------:|:-------:|:-------:|\n|  Utokyo1 | 994 | 794| 200  |\n|  Utokyo2 | 120 | 0 | 120  |\n|  NAU1 | 20 | 0| 20  |\n|  UQ1 | 142 | 0| 142  |\n\nAgain, this is my guess, I have 0 proof!!",
      "votes": null
    },
    {
      "id": "872972",
      "postDate": "06/03/2020 16:53:29",
      "content": "<p>Interesting. That could make for a pretty big shakeup</p>",
      "rawMarkdown": "Interesting. That could make for a pretty big shakeup",
      "votes": null
    },
    {
      "id": "873107",
      "postDate": "06/03/2020 19:51:05",
      "content": "<p>Cool research, Peter!</p>",
      "rawMarkdown": "Cool research, Peter!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 872939,
      "author_name": "pestipeti",
      "author_url": "",
      "post_date": "06/03/2020 16:04:38",
      "content": "<p>From the leaderboard page:</p>\n\n<blockquote>\n  <p>This leaderboard is calculated with approximately 62% of the test data.\n  The final results will be based on the other 38%, so the final standings may be different.</p>\n</blockquote>\n\n<p>62% is a weird number for splitting data. Here is my calculation (the overall number of test samples are from the host's publication, everything else is just my opinion.)</p>\n\n<p>| Source | Overall | Public | Private  |\n|:-------:|:-------:|:-------:|:-------:|\n|  Utokyo1 | 994 | 794| 200  |\n|  Utokyo2 | 120 | 0 | 120  |\n|  NAU1 | 20 | 0| 20  |\n|  UQ1 | 142 | 0| 142  |</p>\n\n<p>Again, this is my guess, I have 0 proof!!</p>",
      "votes": null,
      "replies": [
        {
          "id": 872972,
          "author_name": "alexandersoare",
          "author_url": "",
          "post_date": "06/03/2020 16:53:29",
          "content": "<p>Interesting. That could make for a pretty big shakeup</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 873107,
          "author_name": "sergeyzlobin",
          "author_url": "",
          "post_date": "06/03/2020 19:51:05",
          "content": "<p>Cool research, Peter!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "872898": "I understand that the test data split is 70:30, 70% being for public LB, and 30% for private LB. Does anyone know if the split is stratified across source?\n\n**EDIT** It was pointed out to me that the split is actually 62:38",
    "872939": "From the leaderboard page:\n&gt; This leaderboard is calculated with approximately 62% of the test data.\n&gt; The final results will be based on the other 38%, so the final standings may be different.\n\n62% is a weird number for splitting data. Here is my calculation (the overall number of test samples are from the host's publication, everything else is just my opinion.)\n\n| Source | Overall | Public | Private  |\n|:-------:|:-------:|:-------:|:-------:|\n|  Utokyo1 | 994 | 794| 200  |\n|  Utokyo2 | 120 | 0 | 120  |\n|  NAU1 | 20 | 0| 20  |\n|  UQ1 | 142 | 0| 142  |\n\nAgain, this is my guess, I have 0 proof!!",
    "872972": "Interesting. That could make for a pretty big shakeup",
    "873107": "Cool research, Peter!"
  },
  "source": "meta"
}