{
  "id": 579052,
  "title": "How Many Random Seeds Do You Test",
  "url": "/competitions/birdclef-2025/discussion/579052",
  "author_name": "Ryenhails",
  "post_date": "2025-05-14T20:21:43.031000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>It seems that in both this year's and previous BirdCLEF competitions, random seeds have been a significant source of instability in model performance. </p>\n<p>For the same training setup (assuming you have confidence in your improvements), how many different random seeds do you typically try to explore the variance in results?</p>",
  "messages": [
    {
      "id": 3202092,
      "postDate": "2025-05-14T20:21:43.030Z",
      "content": "<p>It seems that in both this year's and previous BirdCLEF competitions, random seeds have been a significant source of instability in model performance. </p>\n<p>For the same training setup (assuming you have confidence in your improvements), how many different random seeds do you typically try to explore the variance in results?</p>",
      "rawMarkdown": "It seems that in both this year's and previous BirdCLEF competitions, random seeds have been a significant source of instability in model performance. \n\nFor the same training setup (assuming you have confidence in your improvements), how many different random seeds do you typically try to explore the variance in results?",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3202092": "It seems that in both this year's and previous BirdCLEF competitions, random seeds have been a significant source of instability in model performance. \n\nFor the same training setup (assuming you have confidence in your improvements), how many different random seeds do you typically try to explore the variance in results?"
  }
}