{
  "id": 232351,
  "title": "Training Models Outside Kaggle",
  "url": "/competitions/coleridgeinitiative-show-us-the-data/discussion/232351",
  "author_name": "",
  "post_date": "2021-04-13T11:30:11.164420600Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>This competition takes submissions from notebooks which match certain qualifications. One such qualification being it should run in 9 hours.  Also, the \"code requirement\" section notes the following:</p>\n<blockquote>\n  <p>Freely &amp; publicly available external data is allowed, including pre-trained models</p>\n</blockquote>\n<p>Does that mean we can train models elsewhere (e.g. previous versions of notebooks, training on GCP etc.) and then just use these to make prediction within the final notebook which is allowed to run for 9 hours?</p>",
  "messages": [
    {
      "id": "1272275",
      "postDate": "04/13/2021 11:30:11",
      "content": "<p>This competition takes submissions from notebooks which match certain qualifications. One such qualification being it should run in 9 hours.  Also, the \"code requirement\" section notes the following:</p>\n<blockquote>\n  <p>Freely &amp; publicly available external data is allowed, including pre-trained models</p>\n</blockquote>\n<p>Does that mean we can train models elsewhere (e.g. previous versions of notebooks, training on GCP etc.) and then just use these to make prediction within the final notebook which is allowed to run for 9 hours?</p>",
      "rawMarkdown": "This competition takes submissions from notebooks which match certain qualifications. One such qualification being it should run in 9 hours.  Also, the \"code requirement\" section notes the following:\n\n> Freely & publicly available external data is allowed, including pre-trained models\n\nDoes that mean we can train models elsewhere (e.g. previous versions of notebooks, training on GCP etc.) and then just use these to make prediction within the final notebook which is allowed to run for 9 hours?",
      "votes": null
    },
    {
      "id": "1273120",
      "postDate": "04/14/2021 05:25:26",
      "content": "<p>Yes <a href=\"https://www.kaggle.com/anubhav3377\" target=\"_blank\">@anubhav3377</a> <br>\nYou can approach it in a 2-step manner like a training notebook and a separate inference notebook that used the saved model from the training notebook. This is a pretty standard practice for most of the code competitions. It also allows you to use resources like TPU and internet on the training notebook as well, which are otherwise not allowed in code competitions.<br>\nAll the best! 😊</p>",
      "rawMarkdown": "Yes @anubhav3377 \nYou can approach it in a 2-step manner like a training notebook and a separate inference notebook that used the saved model from the training notebook. This is a pretty standard practice for most of the code competitions. It also allows you to use resources like TPU and internet on the training notebook as well, which are otherwise not allowed in code competitions.\nAll the best! 😊",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1273120,
      "author_name": "manabendrarout",
      "author_url": "",
      "post_date": "04/14/2021 05:25:26",
      "content": "<p>Yes <a href=\"https://www.kaggle.com/anubhav3377\" target=\"_blank\">@anubhav3377</a> <br>\nYou can approach it in a 2-step manner like a training notebook and a separate inference notebook that used the saved model from the training notebook. This is a pretty standard practice for most of the code competitions. It also allows you to use resources like TPU and internet on the training notebook as well, which are otherwise not allowed in code competitions.<br>\nAll the best! 😊</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1272275": "This competition takes submissions from notebooks which match certain qualifications. One such qualification being it should run in 9 hours.  Also, the \"code requirement\" section notes the following:\n\n> Freely & publicly available external data is allowed, including pre-trained models\n\nDoes that mean we can train models elsewhere (e.g. previous versions of notebooks, training on GCP etc.) and then just use these to make prediction within the final notebook which is allowed to run for 9 hours?",
    "1273120": "Yes @anubhav3377 \nYou can approach it in a 2-step manner like a training notebook and a separate inference notebook that used the saved model from the training notebook. This is a pretty standard practice for most of the code competitions. It also allows you to use resources like TPU and internet on the training notebook as well, which are otherwise not allowed in code competitions.\nAll the best! 😊"
  },
  "source": "meta"
}