{
  "id": 263347,
  "title": "Questions regarding submission of notebook",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/263347",
  "author_name": "",
  "post_date": "2021-08-09T05:57:01.425087900Z",
  "votes": 2,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hi all,</p>\n<p>this is my first time participating in a Kaggle competition and I have some questions about what is and isn't allowed in this competition. Any answers would be greatly appreciated.</p>\n<p>First of all, I would like to preprocess the data in my local machine, and upload it as dataset in Kaggle, and use that for training. Also, I want to upload pretrained model weights as dataset and use it in my notebook. According to the competition rules, I am allowed to use \"external datset\" as long as it is accessible to all participants. So am I allowed to upload my preprocessed data and pretrained model weights for using in my submission if I make them public?</p>\n<p>Also, does the notebook have to contain training? If the use of pretrained models is allowed, couldn't we simply train the model for a long time offline and use it in the notebook instead of training the model in the notebook? I suppose the point of code competition is to provide all participants with same computing resources for fair competition, but wouldn't using models pretrained elsewhere for submission make the competition unfair?</p>\n<p>Thank you all in advance.</p>",
  "messages": [
    {
      "id": "1460894",
      "postDate": "08/09/2021 05:57:01",
      "content": "<p>Hi all,</p>\n<p>this is my first time participating in a Kaggle competition and I have some questions about what is and isn't allowed in this competition. Any answers would be greatly appreciated.</p>\n<p>First of all, I would like to preprocess the data in my local machine, and upload it as dataset in Kaggle, and use that for training. Also, I want to upload pretrained model weights as dataset and use it in my notebook. According to the competition rules, I am allowed to use \"external datset\" as long as it is accessible to all participants. So am I allowed to upload my preprocessed data and pretrained model weights for using in my submission if I make them public?</p>\n<p>Also, does the notebook have to contain training? If the use of pretrained models is allowed, couldn't we simply train the model for a long time offline and use it in the notebook instead of training the model in the notebook? I suppose the point of code competition is to provide all participants with same computing resources for fair competition, but wouldn't using models pretrained elsewhere for submission make the competition unfair?</p>\n<p>Thank you all in advance.</p>",
      "rawMarkdown": "Hi all,\n\nthis is my first time participating in a Kaggle competition and I have some questions about what is and isn't allowed in this competition. Any answers would be greatly appreciated.\n\nFirst of all, I would like to preprocess the data in my local machine, and upload it as dataset in Kaggle, and use that for training. Also, I want to upload pretrained model weights as dataset and use it in my notebook. According to the competition rules, I am allowed to use \"external datset\" as long as it is accessible to all participants. So am I allowed to upload my preprocessed data and pretrained model weights for using in my submission if I make them public?\n\nAlso, does the notebook have to contain training? If the use of pretrained models is allowed, couldn't we simply train the model for a long time offline and use it in the notebook instead of training the model in the notebook? I suppose the point of code competition is to provide all participants with same computing resources for fair competition, but wouldn't using models pretrained elsewhere for submission make the competition unfair?\n\nThank you all in advance.",
      "votes": null
    },
    {
      "id": "1462187",
      "postDate": "08/09/2021 18:07:04",
      "content": "<p><code>Also, does the notebook have to contain training? If the use of pretrained models is allowed, couldn't we simply train the model for a long time offline and use it in the notebook instead of training the model in the notebook? I suppose the point of code competition is to provide all participants with same computing resources for fair competition, but wouldn't using models pretrained elsewhere for submission make the competition unfair?</code></p>\n<p>Yes this what people usually do … you train model locally and then upload to Kaggle as Dataset. When creating submission you can attach path to your <code>Dataset</code> that contain model and use it to run infrence</p>",
      "rawMarkdown": "`Also, does the notebook have to contain training? If the use of pretrained models is allowed, couldn't we simply train the model for a long time offline and use it in the notebook instead of training the model in the notebook? I suppose the point of code competition is to provide all participants with same computing resources for fair competition, but wouldn't using models pretrained elsewhere for submission make the competition unfair?`\n\nYes this what people usually do ... you train model locally and then upload to Kaggle as Dataset. When creating submission you can attach path to your `Dataset` that contain model and use it to run infrence",
      "votes": null
    },
    {
      "id": "1462233",
      "postDate": "08/09/2021 18:39:54",
      "content": "<p>Assuming you only use the competition data for your training, you do not need to make your personal preprocessed data and personal model public. </p>\n<p>You can use any resources you want for training. It's only the inference portion that must run in a timed notebook.</p>\n<p>-Rich</p>",
      "rawMarkdown": "Assuming you only use the competition data for your training, you do not need to make your personal preprocessed data and personal model public. \n\nYou can use any resources you want for training. It's only the inference portion that must run in a timed notebook.\n\n-Rich",
      "votes": null
    },
    {
      "id": "1462807",
      "postDate": "08/10/2021 02:16:15",
      "content": "<p>I see, thats how it usually works in Kaggle. Thanks for the reply!</p>",
      "rawMarkdown": "I see, thats how it usually works in Kaggle. Thanks for the reply!",
      "votes": null
    },
    {
      "id": "1462819",
      "postDate": "08/10/2021 02:24:00",
      "content": "<p>Hi Rich,</p>\n<p>Thanks for the reply! I want to use pretrained models whise weights are obtained from somewhere else (publically available on the internet, but not trained by me and not trained on the comp dataset) and finetune the model using the comp dataset. Does it mean then that I have to make the pretrained models' weights public on Kaggle?</p>",
      "rawMarkdown": "Hi Rich,\n\nThanks for the reply! I want to use pretrained models whise weights are obtained from somewhere else (publically available on the internet, but not trained by me and not trained on the comp dataset) and finetune the model using the comp dataset. Does it mean then that I have to make the pretrained models' weights public on Kaggle?",
      "votes": null
    },
    {
      "id": "1464577",
      "postDate": "08/10/2021 16:12:24",
      "content": "<p>The rules for external data:</p>\n<p>C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants. The ability to use External Data under this Section 7.C (External Data) does not limit your other obligations under these Competition Rules, including but not limited to Section 11 (Winners Obligations).</p>\n<p>In older competitions, there was often a pinned discussion where people posted the link to external data. I believe that is no longer required. So, you don't have to point out where you are getting the data to competitors, but it needs to be publicly available at no cost.</p>\n<p>-Rich</p>",
      "rawMarkdown": "The rules for external data:\n\nC. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants. The ability to use External Data under this Section 7.C (External Data) does not limit your other obligations under these Competition Rules, including but not limited to Section 11 (Winners Obligations).\n\nIn older competitions, there was often a pinned discussion where people posted the link to external data. I believe that is no longer required. So, you don't have to point out where you are getting the data to competitors, but it needs to be publicly available at no cost.\n\n-Rich",
      "votes": null
    },
    {
      "id": "1465214",
      "postDate": "08/11/2021 00:02:31",
      "content": "<p>Got it. Thanks a lot!</p>",
      "rawMarkdown": "Got it. Thanks a lot!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1462187,
      "author_name": "drhabib",
      "author_url": "",
      "post_date": "08/09/2021 18:07:04",
      "content": "<p><code>Also, does the notebook have to contain training? If the use of pretrained models is allowed, couldn't we simply train the model for a long time offline and use it in the notebook instead of training the model in the notebook? I suppose the point of code competition is to provide all participants with same computing resources for fair competition, but wouldn't using models pretrained elsewhere for submission make the competition unfair?</code></p>\n<p>Yes this what people usually do … you train model locally and then upload to Kaggle as Dataset. When creating submission you can attach path to your <code>Dataset</code> that contain model and use it to run infrence</p>",
      "votes": null,
      "replies": [
        {
          "id": 1462807,
          "author_name": "jinuahn",
          "author_url": "",
          "post_date": "08/10/2021 02:16:15",
          "content": "<p>I see, thats how it usually works in Kaggle. Thanks for the reply!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1462233,
      "author_name": "richardepstein",
      "author_url": "",
      "post_date": "08/09/2021 18:39:54",
      "content": "<p>Assuming you only use the competition data for your training, you do not need to make your personal preprocessed data and personal model public. </p>\n<p>You can use any resources you want for training. It's only the inference portion that must run in a timed notebook.</p>\n<p>-Rich</p>",
      "votes": null,
      "replies": [
        {
          "id": 1462819,
          "author_name": "jinuahn",
          "author_url": "",
          "post_date": "08/10/2021 02:24:00",
          "content": "<p>Hi Rich,</p>\n<p>Thanks for the reply! I want to use pretrained models whise weights are obtained from somewhere else (publically available on the internet, but not trained by me and not trained on the comp dataset) and finetune the model using the comp dataset. Does it mean then that I have to make the pretrained models' weights public on Kaggle?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1464577,
      "author_name": "richardepstein",
      "author_url": "",
      "post_date": "08/10/2021 16:12:24",
      "content": "<p>The rules for external data:</p>\n<p>C. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants. The ability to use External Data under this Section 7.C (External Data) does not limit your other obligations under these Competition Rules, including but not limited to Section 11 (Winners Obligations).</p>\n<p>In older competitions, there was often a pinned discussion where people posted the link to external data. I believe that is no longer required. So, you don't have to point out where you are getting the data to competitors, but it needs to be publicly available at no cost.</p>\n<p>-Rich</p>",
      "votes": null,
      "replies": [
        {
          "id": 1465214,
          "author_name": "jinuahn",
          "author_url": "",
          "post_date": "08/11/2021 00:02:31",
          "content": "<p>Got it. Thanks a lot!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1460894": "Hi all,\n\nthis is my first time participating in a Kaggle competition and I have some questions about what is and isn't allowed in this competition. Any answers would be greatly appreciated.\n\nFirst of all, I would like to preprocess the data in my local machine, and upload it as dataset in Kaggle, and use that for training. Also, I want to upload pretrained model weights as dataset and use it in my notebook. According to the competition rules, I am allowed to use \"external datset\" as long as it is accessible to all participants. So am I allowed to upload my preprocessed data and pretrained model weights for using in my submission if I make them public?\n\nAlso, does the notebook have to contain training? If the use of pretrained models is allowed, couldn't we simply train the model for a long time offline and use it in the notebook instead of training the model in the notebook? I suppose the point of code competition is to provide all participants with same computing resources for fair competition, but wouldn't using models pretrained elsewhere for submission make the competition unfair?\n\nThank you all in advance.",
    "1462187": "`Also, does the notebook have to contain training? If the use of pretrained models is allowed, couldn't we simply train the model for a long time offline and use it in the notebook instead of training the model in the notebook? I suppose the point of code competition is to provide all participants with same computing resources for fair competition, but wouldn't using models pretrained elsewhere for submission make the competition unfair?`\n\nYes this what people usually do ... you train model locally and then upload to Kaggle as Dataset. When creating submission you can attach path to your `Dataset` that contain model and use it to run infrence",
    "1462233": "Assuming you only use the competition data for your training, you do not need to make your personal preprocessed data and personal model public. \n\nYou can use any resources you want for training. It's only the inference portion that must run in a timed notebook.\n\n-Rich",
    "1462807": "I see, thats how it usually works in Kaggle. Thanks for the reply!",
    "1462819": "Hi Rich,\n\nThanks for the reply! I want to use pretrained models whise weights are obtained from somewhere else (publically available on the internet, but not trained by me and not trained on the comp dataset) and finetune the model using the comp dataset. Does it mean then that I have to make the pretrained models' weights public on Kaggle?",
    "1464577": "The rules for external data:\n\nC. External Data. You may use data other than the Competition Data (“External Data”) to develop and test your Submissions. However, you will ensure the External Data is publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants. The ability to use External Data under this Section 7.C (External Data) does not limit your other obligations under these Competition Rules, including but not limited to Section 11 (Winners Obligations).\n\nIn older competitions, there was often a pinned discussion where people posted the link to external data. I believe that is no longer required. So, you don't have to point out where you are getting the data to competitors, but it needs to be publicly available at no cost.\n\n-Rich",
    "1465214": "Got it. Thanks a lot!"
  },
  "source": "meta"
}