{
  "id": 371620,
  "title": "Pre-trained models limitations",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/371620",
  "author_name": "Daria Nogina",
  "post_date": "2022-12-11T10:41:46.177000",
  "votes": 3,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Dear community,<br>\nIt's my first code competition on Kaggle, therefore my question may be naive. I wonder if I can pretrain my model on the competition data using my resources and then submit only the inference notebook with preloaded weights? I think it agrees with the competition rules which state that competitors are allowed to use pre-trained models.</p>",
  "messages": [
    {
      "id": 2061778,
      "postDate": "2022-12-11T13:29:27.403Z",
      "content": "<p>Hey <a href=\"https://www.kaggle.com/diotima\" target=\"_blank\">@diotima</a>!</p>\n<p>What you describe is how most competitors approach the problem. They train the model on their own machine and upload the weights.</p>\n<p>Many public notebooks take a similar approach.</p>\n<p>Generally, when pre-trained weights are allowed, what is meant that you can use weights for models pre-trained on Imagenet or some other publicly available dataset. Still, your idea on how to approach the competition is most certainly within the competition rules 🙂</p>\n<p>Happy Kaggling and best of luck in the challenge!</p>",
      "rawMarkdown": "Hey @diotima!\n\nWhat you describe is how most competitors approach the problem. They train the model on their own machine and upload the weights.\n\nMany public notebooks take a similar approach.\n\nGenerally, when pre-trained weights are allowed, what is meant that you can use weights for models pre-trained on Imagenet or some other publicly available dataset. Still, your idea on how to approach the competition is most certainly within the competition rules 🙂\n\nHappy Kaggling and best of luck in the challenge!",
      "votes": 3
    },
    {
      "id": 2061629,
      "postDate": "2022-12-11T10:41:46.177Z",
      "content": "<p>Dear community,<br>\nIt's my first code competition on Kaggle, therefore my question may be naive. I wonder if I can pretrain my model on the competition data using my resources and then submit only the inference notebook with preloaded weights? I think it agrees with the competition rules which state that competitors are allowed to use pre-trained models.</p>",
      "rawMarkdown": "Dear community,\nIt's my first code competition on Kaggle, therefore my question may be naive. I wonder if I can pretrain my model on the competition data using my resources and then submit only the inference notebook with preloaded weights? I think it agrees with the competition rules which state that competitors are allowed to use pre-trained models.",
      "votes": 3
    }
  ],
  "comments": [
    {
      "id": 2061778,
      "author_name": "Radek Osmulski",
      "author_url": "",
      "post_date": "2022-12-11T13:29:27.403000",
      "content": "<p>Hey <a href=\"https://www.kaggle.com/diotima\" target=\"_blank\">@diotima</a>!</p>\n<p>What you describe is how most competitors approach the problem. They train the model on their own machine and upload the weights.</p>\n<p>Many public notebooks take a similar approach.</p>\n<p>Generally, when pre-trained weights are allowed, what is meant that you can use weights for models pre-trained on Imagenet or some other publicly available dataset. Still, your idea on how to approach the competition is most certainly within the competition rules 🙂</p>\n<p>Happy Kaggling and best of luck in the challenge!</p>",
      "votes": 3,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2061778": "Hey @diotima!\n\nWhat you describe is how most competitors approach the problem. They train the model on their own machine and upload the weights.\n\nMany public notebooks take a similar approach.\n\nGenerally, when pre-trained weights are allowed, what is meant that you can use weights for models pre-trained on Imagenet or some other publicly available dataset. Still, your idea on how to approach the competition is most certainly within the competition rules 🙂\n\nHappy Kaggling and best of luck in the challenge!",
    "2061629": "Dear community,\nIt's my first code competition on Kaggle, therefore my question may be naive. I wonder if I can pretrain my model on the competition data using my resources and then submit only the inference notebook with preloaded weights? I think it agrees with the competition rules which state that competitors are allowed to use pre-trained models."
  }
}