{
  "id": 580178,
  "title": "What strategies of LB shake downs",
  "url": "/competitions/birdclef-2025/discussion/580178",
  "author_name": "",
  "post_date": "2025-05-23T01:01:38.417770500Z",
  "votes": null,
  "comment_count": 6,
  "views": 0,
  "content": "<p>It’s widely known that BirdCLEF tends to have significant leaderboard shake downs every year. What strategies do you use to deal with this? </p>",
  "messages": [
    {
      "id": "3207567",
      "postDate": "05/23/2025 01:01:38",
      "content": "<p>It’s widely known that BirdCLEF tends to have significant leaderboard shake downs every year. What strategies do you use to deal with this? </p>",
      "rawMarkdown": "It’s widely known that BirdCLEF tends to have significant leaderboard shake downs every year. What strategies do you use to deal with this?",
      "votes": null
    },
    {
      "id": "3207568",
      "postDate": "05/23/2025 01:03:33",
      "content": "<p>pray to the god of shake up.</p>\n<p>The expected shake up range is +-0.02 for top teams, if I estimated it correctly.</p>",
      "rawMarkdown": "pray to the god of shake up.\n\nThe expected shake up range is +-0.02 for top teams, if I estimated it correctly.",
      "votes": null
    },
    {
      "id": "3207810",
      "postDate": "05/23/2025 08:57:07",
      "content": "<p>May I ask how you estimated this value? based on pervious BirdCLEF competitions shake ups or? </p>",
      "rawMarkdown": "May I ask how you estimated this value? based on pervious BirdCLEF competitions shake ups or?",
      "votes": null
    },
    {
      "id": "3207892",
      "postDate": "05/23/2025 11:38:51",
      "content": "<p>I can tell from my experiments that the shake up will not be huge, but I could be wrong.</p>",
      "rawMarkdown": "I can tell from my experiments that the shake up will not be huge, but I could be wrong.",
      "votes": null
    },
    {
      "id": "3208059",
      "postDate": "05/23/2025 16:23:35",
      "content": "<p>yes, also most public notebooks are focusing on post processing which is kinda overfitting on LB.</p>",
      "rawMarkdown": "yes, also most public notebooks are focusing on post processing which is kinda overfitting on LB.",
      "votes": null
    },
    {
      "id": "3209678",
      "postDate": "05/26/2025 06:52:45",
      "content": "<p>I simply applied a trick to my single model, which originally had an LB score of 0.832, and its LB jumped to 0.869. I can’t help but suspect this is just overfitting to the LB.</p>",
      "rawMarkdown": "I simply applied a trick to my single model, which originally had an LB score of 0.832, and its LB jumped to 0.869. I can’t help but suspect this is just overfitting to the LB.",
      "votes": null
    },
    {
      "id": "3209699",
      "postDate": "05/26/2025 07:26:36",
      "content": "<p>Amazing trick. Hope you can introduce it after the competition even if it maybe not work on private datasets.</p>",
      "rawMarkdown": "Amazing trick. Hope you can introduce it after the competition even if it maybe not work on private datasets.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3207568,
      "author_name": "yuanzhezhou",
      "author_url": "",
      "post_date": "05/23/2025 01:03:33",
      "content": "<p>pray to the god of shake up.</p>\n<p>The expected shake up range is +-0.02 for top teams, if I estimated it correctly.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3207810,
          "author_name": "mahdiseddigh",
          "author_url": "",
          "post_date": "05/23/2025 08:57:07",
          "content": "<p>May I ask how you estimated this value? based on pervious BirdCLEF competitions shake ups or? </p>",
          "votes": null,
          "replies": [
            {
              "id": 3207892,
              "author_name": "yuanzhezhou",
              "author_url": "",
              "post_date": "05/23/2025 11:38:51",
              "content": "<p>I can tell from my experiments that the shake up will not be huge, but I could be wrong.</p>",
              "votes": null,
              "replies": [
                {
                  "id": 3208059,
                  "author_name": "mahdiseddigh",
                  "author_url": "",
                  "post_date": "05/23/2025 16:23:35",
                  "content": "<p>yes, also most public notebooks are focusing on post processing which is kinda overfitting on LB.</p>",
                  "votes": null,
                  "replies": []
                }
              ]
            }
          ]
        }
      ]
    },
    {
      "id": 3209678,
      "author_name": "sorawww31",
      "author_url": "",
      "post_date": "05/26/2025 06:52:45",
      "content": "<p>I simply applied a trick to my single model, which originally had an LB score of 0.832, and its LB jumped to 0.869. I can’t help but suspect this is just overfitting to the LB.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3209699,
          "author_name": "kurisew",
          "author_url": "",
          "post_date": "05/26/2025 07:26:36",
          "content": "<p>Amazing trick. Hope you can introduce it after the competition even if it maybe not work on private datasets.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3207567": "It’s widely known that BirdCLEF tends to have significant leaderboard shake downs every year. What strategies do you use to deal with this?",
    "3207568": "pray to the god of shake up.\n\nThe expected shake up range is +-0.02 for top teams, if I estimated it correctly.",
    "3207810": "May I ask how you estimated this value? based on pervious BirdCLEF competitions shake ups or?",
    "3207892": "I can tell from my experiments that the shake up will not be huge, but I could be wrong.",
    "3208059": "yes, also most public notebooks are focusing on post processing which is kinda overfitting on LB.",
    "3209678": "I simply applied a trick to my single model, which originally had an LB score of 0.832, and its LB jumped to 0.869. I can’t help but suspect this is just overfitting to the LB.",
    "3209699": "Amazing trick. Hope you can introduce it after the competition even if it maybe not work on private datasets."
  },
  "source": "meta"
}