{
  "id": 399161,
  "title": "Is it allowed to train a model off line and upload it (predict ready)",
  "url": "/competitions/tlvmc-parkinsons-freezing-gait-prediction/discussion/399161",
  "author_name": "",
  "post_date": "2023-04-02T19:19:56.479623600Z",
  "votes": 6,
  "comment_count": 8,
  "views": 0,
  "content": "<p>I was wondering, is it ok to train a model offline - save it (model.save()) and load it for predictions?<br>\nThe reason I'm not sure is because of the computation limits.<br>\nIs training offline follow this competition's rules?</p>",
  "messages": [
    {
      "id": "2206717",
      "postDate": "04/02/2023 19:19:56",
      "content": "<p>I was wondering, is it ok to train a model offline - save it (model.save()) and load it for predictions?<br>\nThe reason I'm not sure is because of the computation limits.<br>\nIs training offline follow this competition's rules?</p>",
      "rawMarkdown": "I was wondering, is it ok to train a model offline - save it (model.save()) and load it for predictions?\nThe reason I'm not sure is because of the computation limits.\nIs training offline follow this competition's rules?",
      "votes": null
    },
    {
      "id": "2206954",
      "postDate": "04/03/2023 03:07:31",
      "content": "<p>You can try it. But subject and meta data of hidden test set can only be available during the submission scoring. So you may lose these infomation in offline training</p>",
      "rawMarkdown": "You can try it. But subject and meta data of hidden test set can only be available during the submission scoring. So you may lose these infomation in offline training",
      "votes": null
    },
    {
      "id": "2207085",
      "postDate": "04/03/2023 05:25:09",
      "content": "<p>Interesting, Thank you!</p>",
      "rawMarkdown": "Interesting, Thank you!",
      "votes": null
    },
    {
      "id": "2210716",
      "postDate": "04/05/2023 15:17:00",
      "content": "<p>Yes, I am pretty sure it is allowed and is even a general Kaggle strategy. Offline training can occur in other notebooks (e.g. training using TPU notebook and submission with GPU notebook) or in your local machine.</p>\n<p>You can the upload the trained to a dataset (see <a href=\"https://www.kaggle.com/datasets\" target=\"_blank\">Dataset Homepage</a>). An option for other saved Kaggle Notebooks versions is saving the outputs directly as a dataset or using the output directly in the submission notebook.</p>\n<p>The competition rules just state that for eligibility for prizes one must show the work done (open source the code) to achieve the final results.</p>",
      "rawMarkdown": "Yes, I am pretty sure it is allowed and is even a general Kaggle strategy. Offline training can occur in other notebooks (e.g. training using TPU notebook and submission with GPU notebook) or in your local machine.\n\nYou can the upload the trained to a dataset (see [Dataset Homepage](https://www.kaggle.com/datasets)). An option for other saved Kaggle Notebooks versions is saving the outputs directly as a dataset or using the output directly in the submission notebook.\n\nThe competition rules just state that for eligibility for prizes one must show the work done (open source the code) to achieve the final results.",
      "votes": null
    },
    {
      "id": "2214714",
      "postDate": "04/08/2023 17:25:45",
      "content": "<p>Hidden samples are only for testing, not for training!</p>",
      "rawMarkdown": "Hidden samples are only for testing, not for training!",
      "votes": null
    },
    {
      "id": "2215065",
      "postDate": "04/09/2023 04:40:31",
      "content": "<p>You don’t need to got the labels, unsupervised method such as kmeans is also useful</p>",
      "rawMarkdown": "You don’t need to got the labels, unsupervised method such as kmeans is also useful",
      "votes": null
    },
    {
      "id": "2273208",
      "postDate": "05/25/2023 04:35:40",
      "content": "<p>If a locally trained model is uploaded for use, does the model need to be publicly available?</p>",
      "rawMarkdown": "If a locally trained model is uploaded for use, does the model need to be publicly available?",
      "votes": null
    },
    {
      "id": "2273242",
      "postDate": "05/25/2023 04:55:44",
      "content": "<p>I think not</p>",
      "rawMarkdown": "I think not",
      "votes": null
    },
    {
      "id": "2273389",
      "postDate": "05/25/2023 06:24:18",
      "content": "<p>I see, thank you!</p>",
      "rawMarkdown": "I see, thank you!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2206954,
      "author_name": "xzj19013742",
      "author_url": "",
      "post_date": "04/03/2023 03:07:31",
      "content": "<p>You can try it. But subject and meta data of hidden test set can only be available during the submission scoring. So you may lose these infomation in offline training</p>",
      "votes": null,
      "replies": [
        {
          "id": 2207085,
          "author_name": "avivlevi815",
          "author_url": "",
          "post_date": "04/03/2023 05:25:09",
          "content": "<p>Interesting, Thank you!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 2214714,
          "author_name": "albertoannoni",
          "author_url": "",
          "post_date": "04/08/2023 17:25:45",
          "content": "<p>Hidden samples are only for testing, not for training!</p>",
          "votes": null,
          "replies": [
            {
              "id": 2215065,
              "author_name": "xzj19013742",
              "author_url": "",
              "post_date": "04/09/2023 04:40:31",
              "content": "<p>You don’t need to got the labels, unsupervised method such as kmeans is also useful</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2210716,
      "author_name": "coderrkj",
      "author_url": "",
      "post_date": "04/05/2023 15:17:00",
      "content": "<p>Yes, I am pretty sure it is allowed and is even a general Kaggle strategy. Offline training can occur in other notebooks (e.g. training using TPU notebook and submission with GPU notebook) or in your local machine.</p>\n<p>You can the upload the trained to a dataset (see <a href=\"https://www.kaggle.com/datasets\" target=\"_blank\">Dataset Homepage</a>). An option for other saved Kaggle Notebooks versions is saving the outputs directly as a dataset or using the output directly in the submission notebook.</p>\n<p>The competition rules just state that for eligibility for prizes one must show the work done (open source the code) to achieve the final results.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2273208,
      "author_name": "tomokooo",
      "author_url": "",
      "post_date": "05/25/2023 04:35:40",
      "content": "<p>If a locally trained model is uploaded for use, does the model need to be publicly available?</p>",
      "votes": null,
      "replies": [
        {
          "id": 2273242,
          "author_name": "avivlevi815",
          "author_url": "",
          "post_date": "05/25/2023 04:55:44",
          "content": "<p>I think not</p>",
          "votes": null,
          "replies": [
            {
              "id": 2273389,
              "author_name": "tomokooo",
              "author_url": "",
              "post_date": "05/25/2023 06:24:18",
              "content": "<p>I see, thank you!</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2206717": "I was wondering, is it ok to train a model offline - save it (model.save()) and load it for predictions?\nThe reason I'm not sure is because of the computation limits.\nIs training offline follow this competition's rules?",
    "2206954": "You can try it. But subject and meta data of hidden test set can only be available during the submission scoring. So you may lose these infomation in offline training",
    "2207085": "Interesting, Thank you!",
    "2210716": "Yes, I am pretty sure it is allowed and is even a general Kaggle strategy. Offline training can occur in other notebooks (e.g. training using TPU notebook and submission with GPU notebook) or in your local machine.\n\nYou can the upload the trained to a dataset (see [Dataset Homepage](https://www.kaggle.com/datasets)). An option for other saved Kaggle Notebooks versions is saving the outputs directly as a dataset or using the output directly in the submission notebook.\n\nThe competition rules just state that for eligibility for prizes one must show the work done (open source the code) to achieve the final results.",
    "2214714": "Hidden samples are only for testing, not for training!",
    "2215065": "You don’t need to got the labels, unsupervised method such as kmeans is also useful",
    "2273208": "If a locally trained model is uploaded for use, does the model need to be publicly available?",
    "2273242": "I think not",
    "2273389": "I see, thank you!"
  },
  "source": "meta"
}