{
  "id": 45442,
  "title": "Leader board validation percentage",
  "url": "/competitions/favorita-grocery-sales-forecasting/discussion/45442",
  "author_name": "",
  "post_date": "2017-12-11T13:23:31.700968Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Leaderboard says \"calculated with approximately 31% of the test data.\"</p>\n\n<p>How those \"aprox 31%\" is being got? is it 5/16 (first five days) or random sample from the whole test set?</p>\n\n<p>I am asking because 31% is really suspicious. It's really close to 5/16. I would expect 30% or 33% more for random sample.</p>",
  "messages": [
    {
      "id": "256203",
      "postDate": "12/11/2017 13:23:31",
      "content": "<p>Leaderboard says \"calculated with approximately 31% of the test data.\"</p>\n\n<p>How those \"aprox 31%\" is being got? is it 5/16 (first five days) or random sample from the whole test set?</p>\n\n<p>I am asking because 31% is really suspicious. It's really close to 5/16. I would expect 30% or 33% more for random sample.</p>",
      "rawMarkdown": "Leaderboard says \"calculated with approximately 31% of the test data.\"\n\nHow those \"aprox 31%\" is being got? is it 5/16 (first five days) or random sample from the whole test set?\n\nI am asking because 31% is really suspicious. It's really close to 5/16. I would expect 30% or 33% more for random sample.",
      "votes": null
    },
    {
      "id": "256259",
      "postDate": "12/11/2017 15:49:02",
      "content": "<p>It'd be pretty easy to test if the first five days are the public LB no? Take an old submission turn all the values past 5/16 in to zeros and resubmit. If it's the same score then you're correct.</p>",
      "rawMarkdown": "It'd be pretty easy to test if the first five days are the public LB no? Take an old submission turn all the values past 5/16 in to zeros and resubmit. If it's the same score then you're correct.",
      "votes": null
    },
    {
      "id": "256271",
      "postDate": "12/11/2017 16:19:42",
      "content": "<p>Fair enough, thanks.</p>",
      "rawMarkdown": "Fair enough, thanks.",
      "votes": null
    },
    {
      "id": "261555",
      "postDate": "12/22/2017 23:53:13",
      "content": "<p>...and, as it happens, tarboxl <a href=\"https://www.kaggle.com/tarobxl/how-the-test-set-is-split-lb-0-532\">already did it</a></p>",
      "rawMarkdown": "...and, as it happens, tarboxl [already did it][1]\n\n [1]: https://www.kaggle.com/tarobxl/how-the-test-set-is-split-lb-0-532",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 256259,
      "author_name": "michaelsnell",
      "author_url": "",
      "post_date": "12/11/2017 15:49:02",
      "content": "<p>It'd be pretty easy to test if the first five days are the public LB no? Take an old submission turn all the values past 5/16 in to zeros and resubmit. If it's the same score then you're correct.</p>",
      "votes": null,
      "replies": [
        {
          "id": 256271,
          "author_name": "dchirikov",
          "author_url": "",
          "post_date": "12/11/2017 16:19:42",
          "content": "<p>Fair enough, thanks.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 261555,
          "author_name": "aharless",
          "author_url": "",
          "post_date": "12/22/2017 23:53:13",
          "content": "<p>...and, as it happens, tarboxl <a href=\"https://www.kaggle.com/tarobxl/how-the-test-set-is-split-lb-0-532\">already did it</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "256203": "Leaderboard says \"calculated with approximately 31% of the test data.\"\n\nHow those \"aprox 31%\" is being got? is it 5/16 (first five days) or random sample from the whole test set?\n\nI am asking because 31% is really suspicious. It's really close to 5/16. I would expect 30% or 33% more for random sample.",
    "256259": "It'd be pretty easy to test if the first five days are the public LB no? Take an old submission turn all the values past 5/16 in to zeros and resubmit. If it's the same score then you're correct.",
    "256271": "Fair enough, thanks.",
    "261555": "...and, as it happens, tarboxl [already did it][1]\n\n [1]: https://www.kaggle.com/tarobxl/how-the-test-set-is-split-lb-0-532"
  },
  "source": "meta"
}