{
  "id": 263913,
  "title": "Hidden dataset",
  "url": "/competitions/siim-covid19-detection/discussion/263913",
  "author_name": "Michał Choiński",
  "post_date": "2021-08-10T14:57:31.104000",
  "votes": 5,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Would it be possible to reveal the content of the hidden dataset, or at least a distribution of the classes? I am curious how much it deviated from the public  and train datasets. We had very similar CV scores to the ones visible in the public LB and they significantly dropped in the private LB. It would be great to understand it well if the reason was just that we were optimizing scores of known OOFs + LB, or there was something more behind it. </p>",
  "messages": [
    {
      "id": 1464400,
      "postDate": "2021-08-10T14:57:31.103Z",
      "content": "<p>Would it be possible to reveal the content of the hidden dataset, or at least a distribution of the classes? I am curious how much it deviated from the public  and train datasets. We had very similar CV scores to the ones visible in the public LB and they significantly dropped in the private LB. It would be great to understand it well if the reason was just that we were optimizing scores of known OOFs + LB, or there was something more behind it. </p>",
      "rawMarkdown": "Would it be possible to reveal the content of the hidden dataset, or at least a distribution of the classes? I am curious how much it deviated from the public  and train datasets. We had very similar CV scores to the ones visible in the public LB and they significantly dropped in the private LB. It would be great to understand it well if the reason was just that we were optimizing scores of known OOFs + LB, or there was something more behind it. ",
      "votes": 5
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1464400": "Would it be possible to reveal the content of the hidden dataset, or at least a distribution of the classes? I am curious how much it deviated from the public  and train datasets. We had very similar CV scores to the ones visible in the public LB and they significantly dropped in the private LB. It would be great to understand it well if the reason was just that we were optimizing scores of known OOFs + LB, or there was something more behind it. "
  }
}