{
  "id": 682321,
  "title": "is possible shake up?",
  "url": "/competitions/deep-past-initiative-machine-translation/discussion/682321",
  "author_name": "",
  "post_date": "2026-03-18T05:48:49.076067300Z",
  "votes": -1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>since training samples around ~1.5k with repeated words, but testing samples 4k the leaderboard is calculated with approximately 34% of the test data. The final results will be based on the other 66%.</p>",
  "messages": [
    {
      "id": "3423056",
      "postDate": "03/18/2026 05:48:49",
      "content": "<p>since training samples around ~1.5k with repeated words, but testing samples 4k the leaderboard is calculated with approximately 34% of the test data. The final results will be based on the other 66%.</p>",
      "rawMarkdown": "since training samples around ~1.5k with repeated words, but testing samples 4k the leaderboard is calculated with approximately 34% of the test data. The final results will be based on the other 66%.",
      "votes": null
    },
    {
      "id": "3423251",
      "postDate": "03/18/2026 10:42:50",
      "content": "<p>Yes, as in every competition where training data is from a different distribution than test data (full text vs sentences, possible different preprocessing)</p>",
      "rawMarkdown": "Yes, as in every competition where training data is from a different distribution than test data (full text vs sentences, possible different preprocessing)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3423251,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "03/18/2026 10:42:50",
      "content": "<p>Yes, as in every competition where training data is from a different distribution than test data (full text vs sentences, possible different preprocessing)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3423056": "since training samples around ~1.5k with repeated words, but testing samples 4k the leaderboard is calculated with approximately 34% of the test data. The final results will be based on the other 66%.",
    "3423251": "Yes, as in every competition where training data is from a different distribution than test data (full text vs sentences, possible different preprocessing)"
  },
  "source": "meta"
}