{
  "id": 578314,
  "title": "Is it better for you to use 5-fold cross-validation or train on all the data?",
  "url": "/competitions/birdclef-2025/discussion/578314",
  "author_name": "",
  "post_date": "2025-05-10T02:53:31.234438900Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Currently, the results of 5-fold cross-validation are much better than training on the entire dataset.</p>",
  "messages": [
    {
      "id": "3198808",
      "postDate": "05/10/2025 02:53:31",
      "content": "<p>Currently, the results of 5-fold cross-validation are much better than training on the entire dataset.</p>",
      "rawMarkdown": "Currently, the results of 5-fold cross-validation are much better than training on the entire dataset.",
      "votes": null
    },
    {
      "id": "3199484",
      "postDate": "05/11/2025 04:09:50",
      "content": "<p>For me 5-fold cross-validation are better than training with all the data. But it cost too many time. Even I use ema or swa, my single model (with all data) is always worse than cross-validation. I don't know how to fix it. :(</p>",
      "rawMarkdown": "For me 5-fold cross-validation are better than training with all the data. But it cost too many time. Even I use ema or swa, my single model (with all data) is always worse than cross-validation. I don't know how to fix it. :(",
      "votes": null
    },
    {
      "id": "3199510",
      "postDate": "05/11/2025 05:02:15",
      "content": "<p>K-fold model is not a good choice due to the limited time. It's \"better performance\" is because it's a kind of ensemble. I think using diverse backbones trained on entire data is the right way.</p>",
      "rawMarkdown": "K-fold model is not a good choice due to the limited time. It's \"better performance\" is because it's a kind of ensemble. I think using diverse backbones trained on entire data is the right way.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3199484,
      "author_name": "i2nfinit3y",
      "author_url": "",
      "post_date": "05/11/2025 04:09:50",
      "content": "<p>For me 5-fold cross-validation are better than training with all the data. But it cost too many time. Even I use ema or swa, my single model (with all data) is always worse than cross-validation. I don't know how to fix it. :(</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3199510,
      "author_name": "takanashihumbert",
      "author_url": "",
      "post_date": "05/11/2025 05:02:15",
      "content": "<p>K-fold model is not a good choice due to the limited time. It's \"better performance\" is because it's a kind of ensemble. I think using diverse backbones trained on entire data is the right way.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3198808": "Currently, the results of 5-fold cross-validation are much better than training on the entire dataset.",
    "3199484": "For me 5-fold cross-validation are better than training with all the data. But it cost too many time. Even I use ema or swa, my single model (with all data) is always worse than cross-validation. I don't know how to fix it. :(",
    "3199510": "K-fold model is not a good choice due to the limited time. It's \"better performance\" is because it's a kind of ensemble. I think using diverse backbones trained on entire data is the right way."
  },
  "source": "meta"
}