{
  "id": 440577,
  "title": "Use of pre-trained models as external dataset",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/440577",
  "author_name": "",
  "post_date": "2023-09-15T13:07:07.080596100Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I know it's a silly question but since the usage of external dataset is allowed, can we do this actually?</p>\n<p>Train model locally or somewhere with massive computing power, then save the config/metadata/weights in JSON/CSV format or anything. Upload to Kaggle as a dataset, load it back as a model, then just do the inference without needing time to train it. Boom free 9 hours for submission.</p>",
  "messages": [
    {
      "id": "2440362",
      "postDate": "09/15/2023 13:07:07",
      "content": "<p>I know it's a silly question but since the usage of external dataset is allowed, can we do this actually?</p>\n<p>Train model locally or somewhere with massive computing power, then save the config/metadata/weights in JSON/CSV format or anything. Upload to Kaggle as a dataset, load it back as a model, then just do the inference without needing time to train it. Boom free 9 hours for submission.</p>",
      "rawMarkdown": "I know it's a silly question but since the usage of external dataset is allowed, can we do this actually?\n\nTrain model locally or somewhere with massive computing power, then save the config/metadata/weights in JSON/CSV format or anything. Upload to Kaggle as a dataset, load it back as a model, then just do the inference without needing time to train it. Boom free 9 hours for submission.",
      "votes": null
    },
    {
      "id": "2440377",
      "postDate": "09/15/2023 13:17:02",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/rimbax\" target=\"_blank\">@rimbax</a>. As stated in the Code Requirements “Freely &amp; publicly available external data is allowed, including pre-trained models”, it is possible.</p>",
      "rawMarkdown": "Hi @rimbax. As stated in the Code Requirements “Freely & publicly available external data is allowed, including pre-trained models”, it is possible.",
      "votes": null
    },
    {
      "id": "2440391",
      "postDate": "09/15/2023 13:22:46",
      "content": "<p>Yep agree with that. It doesn't state \"how\" the external data and pretrained models are being used after all, so i think it's possible to use my technique above.</p>",
      "rawMarkdown": "Yep agree with that. It doesn't state \"how\" the external data and pretrained models are being used after all, so i think it's possible to use my technique above.",
      "votes": null
    },
    {
      "id": "2440861",
      "postDate": "09/15/2023 19:24:26",
      "content": "<p>Yes, that is generally allowed. In fact, it is common in competitions to do that as it is often impossible to fit the models on Kaggle in memory/time.</p>",
      "rawMarkdown": "Yes, that is generally allowed. In fact, it is common in competitions to do that as it is often impossible to fit the models on Kaggle in memory/time.",
      "votes": null
    },
    {
      "id": "2441050",
      "postDate": "09/16/2023 01:40:07",
      "content": "<p>Great to hear. I am actually new to a featured competition in Kaggle, so I have no exposure on that. Thank you for sharing!</p>",
      "rawMarkdown": "Great to hear. I am actually new to a featured competition in Kaggle, so I have no exposure on that. Thank you for sharing!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2440377,
      "author_name": "mitsuyasuhoshino",
      "author_url": "",
      "post_date": "09/15/2023 13:17:02",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/rimbax\" target=\"_blank\">@rimbax</a>. As stated in the Code Requirements “Freely &amp; publicly available external data is allowed, including pre-trained models”, it is possible.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2440391,
          "author_name": "rimbax",
          "author_url": "",
          "post_date": "09/15/2023 13:22:46",
          "content": "<p>Yep agree with that. It doesn't state \"how\" the external data and pretrained models are being used after all, so i think it's possible to use my technique above.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2440861,
      "author_name": "fritzcremer",
      "author_url": "",
      "post_date": "09/15/2023 19:24:26",
      "content": "<p>Yes, that is generally allowed. In fact, it is common in competitions to do that as it is often impossible to fit the models on Kaggle in memory/time.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2441050,
          "author_name": "rimbax",
          "author_url": "",
          "post_date": "09/16/2023 01:40:07",
          "content": "<p>Great to hear. I am actually new to a featured competition in Kaggle, so I have no exposure on that. Thank you for sharing!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2440362": "I know it's a silly question but since the usage of external dataset is allowed, can we do this actually?\n\nTrain model locally or somewhere with massive computing power, then save the config/metadata/weights in JSON/CSV format or anything. Upload to Kaggle as a dataset, load it back as a model, then just do the inference without needing time to train it. Boom free 9 hours for submission.",
    "2440377": "Hi @rimbax. As stated in the Code Requirements “Freely & publicly available external data is allowed, including pre-trained models”, it is possible.",
    "2440391": "Yep agree with that. It doesn't state \"how\" the external data and pretrained models are being used after all, so i think it's possible to use my technique above.",
    "2440861": "Yes, that is generally allowed. In fact, it is common in competitions to do that as it is often impossible to fit the models on Kaggle in memory/time.",
    "2441050": "Great to hear. I am actually new to a featured competition in Kaggle, so I have no exposure on that. Thank you for sharing!"
  },
  "source": "meta"
}