{
  "id": 582343,
  "title": "Ensembling does not work ",
  "url": "/competitions/birdclef-2025/discussion/582343",
  "author_name": "Giovanni Marco Dall'Olio",
  "post_date": "2025-05-30T15:51:03.264000",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I've tried an Ensembling approach, to see how it performs.</p>\n<p>I started from the dataset shared by <a href=\"https://www.kaggle.com/code/myso1987/birdclef2025-sound-event-visualisations\" target=\"_blank\">https://www.kaggle.com/code/myso1987/birdclef2025-sound-event-visualisations</a> ( Thank you for sharing it!) and used several strategies to generate a solution. </p>\n<p>This approach however does not work, because the notebooks are re-executed during the evaluation process, using hidden data - thus, the shape of the test file may be different - I am not sure how it is implemented</p>\n<p>Kudos to the organizers for setting up this system, I wish this was used for the Playground competitions as well.</p>",
  "messages": [
    {
      "id": 3213875,
      "postDate": "2025-05-30T15:51:03.263Z",
      "content": "<p>I've tried an Ensembling approach, to see how it performs.</p>\n<p>I started from the dataset shared by <a href=\"https://www.kaggle.com/code/myso1987/birdclef2025-sound-event-visualisations\" target=\"_blank\">https://www.kaggle.com/code/myso1987/birdclef2025-sound-event-visualisations</a> ( Thank you for sharing it!) and used several strategies to generate a solution. </p>\n<p>This approach however does not work, because the notebooks are re-executed during the evaluation process, using hidden data - thus, the shape of the test file may be different - I am not sure how it is implemented</p>\n<p>Kudos to the organizers for setting up this system, I wish this was used for the Playground competitions as well.</p>",
      "rawMarkdown": "I've tried an Ensembling approach, to see how it performs.\n\nI started from the dataset shared by https://www.kaggle.com/code/myso1987/birdclef2025-sound-event-visualisations ( Thank you for sharing it!) and used several strategies to generate a solution. \n\nThis approach however does not work, because the notebooks are re-executed during the evaluation process, using hidden data - thus, the shape of the test file may be different - I am not sure how it is implemented\n\nKudos to the organizers for setting up this system, I wish this was used for the Playground competitions as well.\n\n",
      "votes": 4
    },
    {
      "id": 3216973,
      "postDate": "2025-06-04T11:11:10.050Z",
      "content": "<p>is it true? Late fusion of 3 CNNs: 0.74 F1-score (public LB) → 0.69 (private LB)</p>\n<p>Weighted average of AST + HTS-AT: 0.71 F1-score (public LB) → 0.65 (private LB)</p>",
      "rawMarkdown": "is it true? Late fusion of 3 CNNs: 0.74 F1-score (public LB) → 0.69 (private LB)\n\nWeighted average of AST + HTS-AT: 0.71 F1-score (public LB) → 0.65 (private LB)"
    },
    {
      "id": 3217320,
      "postDate": "2025-06-04T21:14:17.110Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 3216973,
      "author_name": "Kenneth Tuchman",
      "author_url": "",
      "post_date": "2025-06-04T11:11:10.050000",
      "content": "<p>is it true? Late fusion of 3 CNNs: 0.74 F1-score (public LB) → 0.69 (private LB)</p>\n<p>Weighted average of AST + HTS-AT: 0.71 F1-score (public LB) → 0.65 (private LB)</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 3217320,
      "author_name": "",
      "author_url": "",
      "post_date": "2025-06-04T21:14:17.110000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3213875": "I've tried an Ensembling approach, to see how it performs.\n\nI started from the dataset shared by https://www.kaggle.com/code/myso1987/birdclef2025-sound-event-visualisations ( Thank you for sharing it!) and used several strategies to generate a solution. \n\nThis approach however does not work, because the notebooks are re-executed during the evaluation process, using hidden data - thus, the shape of the test file may be different - I am not sure how it is implemented\n\nKudos to the organizers for setting up this system, I wish this was used for the Playground competitions as well.\n\n",
    "3216973": "is it true? Late fusion of 3 CNNs: 0.74 F1-score (public LB) → 0.69 (private LB)\n\nWeighted average of AST + HTS-AT: 0.71 F1-score (public LB) → 0.65 (private LB)",
    "3217320": ""
  }
}