{
  "id": 278007,
  "title": "Novice question on competition rules",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/278007",
  "author_name": "",
  "post_date": "2021-10-12T11:09:40.302211600Z",
  "votes": 3,
  "comment_count": 5,
  "views": 0,
  "content": "<p>My teammates and I are all novices in kaggle competitions. And we can't figure out a particular question.<br>\nDoes one have to train the model from scratch in the submitted notebook? Or can one train a model on one's own machine, upload weights as a dataset, add this dataset as data to the notebook and perform only inference on test set?<br>\nAccording to rules only freely &amp; publicly available external data is allowed, including pre-trained models. So it seems that uploading trained weights is not allowed, however we 're not sure.<br>\nThanks for any help in clarifying this question 🙂</p>",
  "messages": [
    {
      "id": "1542309",
      "postDate": "10/12/2021 11:09:40",
      "content": "<p>My teammates and I are all novices in kaggle competitions. And we can't figure out a particular question.<br>\nDoes one have to train the model from scratch in the submitted notebook? Or can one train a model on one's own machine, upload weights as a dataset, add this dataset as data to the notebook and perform only inference on test set?<br>\nAccording to rules only freely &amp; publicly available external data is allowed, including pre-trained models. So it seems that uploading trained weights is not allowed, however we 're not sure.<br>\nThanks for any help in clarifying this question 🙂</p>",
      "rawMarkdown": "My teammates and I are all novices in kaggle competitions. And we can't figure out a particular question.\nDoes one have to train the model from scratch in the submitted notebook? Or can one train a model on one's own machine, upload weights as a dataset, add this dataset as data to the notebook and perform only inference on test set?\nAccording to rules only freely & publicly available external data is allowed, including pre-trained models. So it seems that uploading trained weights is not allowed, however we 're not sure.\nThanks for any help in clarifying this question 🙂",
      "votes": null
    },
    {
      "id": "1542794",
      "postDate": "10/12/2021 19:56:58",
      "content": "<p>I think it is okay to \"train a model on one's own machine, upload weights as a dataset, add this dataset as data to the notebook and perform only inference on test set\"  I don't think your own models count as external data, and I have done this in previous code competitions.  But perhaps the organizers can weigh in with an official answer.</p>",
      "rawMarkdown": "I think it is okay to \"train a model on one's own machine, upload weights as a dataset, add this dataset as data to the notebook and perform only inference on test set\"  I don't think your own models count as external data, and I have done this in previous code competitions.  But perhaps the organizers can weigh in with an official answer.",
      "votes": null
    },
    {
      "id": "1542849",
      "postDate": "10/12/2021 21:06:43",
      "content": "<p>Thank you for the reply!<br>\nDid you add the weights in those competitions as a public dataset or as a private one?</p>",
      "rawMarkdown": "Thank you for the reply!\nDid you add the weights in those competitions as a public dataset or as a private one?",
      "votes": null
    },
    {
      "id": "1542888",
      "postDate": "10/12/2021 22:51:31",
      "content": "<p>As private only.</p>",
      "rawMarkdown": "As private only.",
      "votes": null
    },
    {
      "id": "1543283",
      "postDate": "10/13/2021 10:10:38",
      "content": "<p>Thank you 🙂</p>",
      "rawMarkdown": "Thank you 🙂",
      "votes": null
    },
    {
      "id": "1543616",
      "postDate": "10/13/2021 17:24:59",
      "content": "<p>how will your model improve if you decide to utilize mri planes (coronal, sagittal , axial) from dicom metadata , how would you even use in your model?</p>",
      "rawMarkdown": "how will your model improve if you decide to utilize mri planes (coronal, sagittal , axial) from dicom metadata , how would you even use in your model?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1542794,
      "author_name": "dslate",
      "author_url": "",
      "post_date": "10/12/2021 19:56:58",
      "content": "<p>I think it is okay to \"train a model on one's own machine, upload weights as a dataset, add this dataset as data to the notebook and perform only inference on test set\"  I don't think your own models count as external data, and I have done this in previous code competitions.  But perhaps the organizers can weigh in with an official answer.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1542849,
          "author_name": "arq007",
          "author_url": "",
          "post_date": "10/12/2021 21:06:43",
          "content": "<p>Thank you for the reply!<br>\nDid you add the weights in those competitions as a public dataset or as a private one?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1542888,
          "author_name": "dslate",
          "author_url": "",
          "post_date": "10/12/2021 22:51:31",
          "content": "<p>As private only.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1543283,
          "author_name": "arq007",
          "author_url": "",
          "post_date": "10/13/2021 10:10:38",
          "content": "<p>Thank you 🙂</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1543616,
      "author_name": "aniarya",
      "author_url": "",
      "post_date": "10/13/2021 17:24:59",
      "content": "<p>how will your model improve if you decide to utilize mri planes (coronal, sagittal , axial) from dicom metadata , how would you even use in your model?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1542309": "My teammates and I are all novices in kaggle competitions. And we can't figure out a particular question.\nDoes one have to train the model from scratch in the submitted notebook? Or can one train a model on one's own machine, upload weights as a dataset, add this dataset as data to the notebook and perform only inference on test set?\nAccording to rules only freely & publicly available external data is allowed, including pre-trained models. So it seems that uploading trained weights is not allowed, however we 're not sure.\nThanks for any help in clarifying this question 🙂",
    "1542794": "I think it is okay to \"train a model on one's own machine, upload weights as a dataset, add this dataset as data to the notebook and perform only inference on test set\"  I don't think your own models count as external data, and I have done this in previous code competitions.  But perhaps the organizers can weigh in with an official answer.",
    "1542849": "Thank you for the reply!\nDid you add the weights in those competitions as a public dataset or as a private one?",
    "1542888": "As private only.",
    "1543283": "Thank you 🙂",
    "1543616": "how will your model improve if you decide to utilize mri planes (coronal, sagittal , axial) from dicom metadata , how would you even use in your model?"
  },
  "source": "meta"
}