{
  "id": 479606,
  "title": "Do your larger models perform better?",
  "url": "/competitions/hms-harmful-brain-activity-classification/discussion/479606",
  "author_name": "",
  "post_date": "2024-02-25T11:00:36.623140Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I tried some resnet models and some efficient models, it turns out that small models are better on cv and lb. <br>\nI don't kown if it's only me or not. Maybe it's because we don't have enough data.</p>",
  "messages": [
    {
      "id": "2667794",
      "postDate": "02/25/2024 11:00:36",
      "content": "<p>I tried some resnet models and some efficient models, it turns out that small models are better on cv and lb. <br>\nI don't kown if it's only me or not. Maybe it's because we don't have enough data.</p>",
      "rawMarkdown": "I tried some resnet models and some efficient models, it turns out that small models are better on cv and lb. \nI don't kown if it's only me or not. Maybe it's because we don't have enough data.",
      "votes": null
    },
    {
      "id": "2667831",
      "postDate": "02/25/2024 11:31:57",
      "content": "<p>Larger models seem indeed to overfit quite quickly, good augmentation might improve that situation a bit but you observed exactly what I did.</p>",
      "rawMarkdown": "Larger models seem indeed to overfit quite quickly, good augmentation might improve that situation a bit but you observed exactly what I did.",
      "votes": null
    },
    {
      "id": "2668060",
      "postDate": "02/25/2024 13:45:11",
      "content": "<p>me too, b0&gt;ResNet50&gt;VIT</p>",
      "rawMarkdown": "me too, b0>ResNet50>VIT",
      "votes": null
    },
    {
      "id": "2668729",
      "postDate": "02/25/2024 20:28:35",
      "content": "<p>In my testing most of my best models are smaller versions of the base. Even with augmentation smaller models have been better for me.</p>",
      "rawMarkdown": "In my testing most of my best models are smaller versions of the base. Even with augmentation smaller models have been better for me.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2667831,
      "author_name": "janbrederecke",
      "author_url": "",
      "post_date": "02/25/2024 11:31:57",
      "content": "<p>Larger models seem indeed to overfit quite quickly, good augmentation might improve that situation a bit but you observed exactly what I did.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2668060,
      "author_name": "chenboluo",
      "author_url": "",
      "post_date": "02/25/2024 13:45:11",
      "content": "<p>me too, b0&gt;ResNet50&gt;VIT</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2668729,
      "author_name": "cody11null",
      "author_url": "",
      "post_date": "02/25/2024 20:28:35",
      "content": "<p>In my testing most of my best models are smaller versions of the base. Even with augmentation smaller models have been better for me.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2667794": "I tried some resnet models and some efficient models, it turns out that small models are better on cv and lb. \nI don't kown if it's only me or not. Maybe it's because we don't have enough data.",
    "2667831": "Larger models seem indeed to overfit quite quickly, good augmentation might improve that situation a bit but you observed exactly what I did.",
    "2668060": "me too, b0>ResNet50>VIT",
    "2668729": "In my testing most of my best models are smaller versions of the base. Even with augmentation smaller models have been better for me."
  },
  "source": "meta"
}