{
  "id": 511113,
  "title": "Running Code Environments for this Competition",
  "url": "/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/511113",
  "author_name": "",
  "post_date": "2024-06-09T07:22:46.822900100Z",
  "votes": 1,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hi everyone,</p>\n<p>I've been participating in the RSNA 2024 Lumbar Spine Degenerative Classification competition and have been encountering frequent issues with exceeding the platform's limits. I was wondering how others are managing their code execution environments. Are you running your code exclusively on Kaggle, or are you using local machines or other platforms to develop and test your models before submitting?</p>\n<p>Any insights or recommendations would be greatly appreciated.</p>\n<p>Thank you!</p>",
  "messages": [
    {
      "id": "2863019",
      "postDate": "06/09/2024 07:22:46",
      "content": "<p>Hi everyone,</p>\n<p>I've been participating in the RSNA 2024 Lumbar Spine Degenerative Classification competition and have been encountering frequent issues with exceeding the platform's limits. I was wondering how others are managing their code execution environments. Are you running your code exclusively on Kaggle, or are you using local machines or other platforms to develop and test your models before submitting?</p>\n<p>Any insights or recommendations would be greatly appreciated.</p>\n<p>Thank you!</p>",
      "rawMarkdown": "Hi everyone,\n\nI've been participating in the RSNA 2024 Lumbar Spine Degenerative Classification competition and have been encountering frequent issues with exceeding the platform's limits. I was wondering how others are managing their code execution environments. Are you running your code exclusively on Kaggle, or are you using local machines or other platforms to develop and test your models before submitting?\n\nAny insights or recommendations would be greatly appreciated.\n\nThank you!",
      "votes": null
    },
    {
      "id": "2863049",
      "postDate": "06/09/2024 07:40:43",
      "content": "<p>You should be able to train the models in any other enviroment (e.g., locally, Colab, cloud VM). But final submission must be done on a Kaggle Notebook with internet turned off. You can upload the trained model weights as a private dataset and load the weights during the model run.</p>\n<p>Me personally, I have not hit the limits of Kaggle for training as of yet, so I am continuing to use it. I am not certain what would be the best external option for this competition, but I would suggest looking into exhausting both GPU and TPU options in Kaggle before trying other alternatives.</p>",
      "rawMarkdown": "You should be able to train the models in any other enviroment (e.g., locally, Colab, cloud VM). But final submission must be done on a Kaggle Notebook with internet turned off. You can upload the trained model weights as a private dataset and load the weights during the model run.\n\nMe personally, I have not hit the limits of Kaggle for training as of yet, so I am continuing to use it. I am not certain what would be the best external option for this competition, but I would suggest looking into exhausting both GPU and TPU options in Kaggle before trying other alternatives.",
      "votes": null
    },
    {
      "id": "2863812",
      "postDate": "06/09/2024 17:43:08",
      "content": "<p>You should be able to train the models in any other environment (e.g., locally, Colab, cloud VM). But the final submission must be done on a Kaggle Notebook with the internet turned off. You can upload the trained model weights as a private dataset and load the weights during the model run.</p>\n<p><a href=\"https://www.kaggle.com/coderrkj\" target=\"_blank\">@coderrkj</a> I think this may be against the rules unless the pre-trained model is made publicly and freely available. 'Freely &amp; publicly available external data is allowed, including pre-trained models'- See the code requirements in the overview page. You can also think of it this way: if private pre-trained models were allowed wouldn't it be unfair to others with less compute resources?  Rules such as CPU Notebook &lt;= 9 hours run-time<br>\nGPU Notebook &lt;= 9 hours run-time would not make much sense if private pre-trained models were allowed, but I am not 100% sure</p>",
      "rawMarkdown": "You should be able to train the models in any other environment (e.g., locally, Colab, cloud VM). But the final submission must be done on a Kaggle Notebook with the internet turned off. You can upload the trained model weights as a private dataset and load the weights during the model run.\n\n@coderrkj I think this may be against the rules unless the pre-trained model is made publicly and freely available. 'Freely & publicly available external data is allowed, including pre-trained models'- See the code requirements in the overview page. You can also think of it this way: if private pre-trained models were allowed wouldn't it be unfair to others with less compute resources?  Rules such as CPU Notebook <= 9 hours run-time\nGPU Notebook <= 9 hours run-time would not make much sense if private pre-trained models were allowed, but I am not 100% sure",
      "votes": null
    },
    {
      "id": "2864161",
      "postDate": "06/10/2024 01:48:58",
      "content": "<p>I have noticed that my code takes seconds to completely run but scoring takes about 9 to 15mins and when I used ResNet18 running took seconds but scoring took 2 hours. But I do not know why</p>",
      "rawMarkdown": "I have noticed that my code takes seconds to completely run but scoring takes about 9 to 15mins and when I used ResNet18 running took seconds but scoring took 2 hours. But I do not know why",
      "votes": null
    },
    {
      "id": "2864300",
      "postDate": "06/10/2024 05:00:51",
      "content": "<p>The number of test cases are greater during submission. You need to account for the runtime of ResNet18 on those cases also in the 2 hours it takes during submission.</p>",
      "rawMarkdown": "The number of test cases are greater during submission. You need to account for the runtime of ResNet18 on those cases also in the 2 hours it takes during submission.",
      "votes": null
    },
    {
      "id": "2864492",
      "postDate": "06/10/2024 07:03:27",
      "content": "<p>Thanks for your advice it is greatly appreciated. I'll keep pursuing the Kaggle platform for now I think. </p>\n<p>The 'no internet' element was another learning curve for me. </p>",
      "rawMarkdown": "Thanks for your advice it is greatly appreciated. I'll keep pursuing the Kaggle platform for now I think. \n\nThe 'no internet' element was another learning curve for me.",
      "votes": null
    },
    {
      "id": "2864493",
      "postDate": "06/10/2024 07:03:52",
      "content": "<p>hmmm good point</p>",
      "rawMarkdown": "hmmm good point",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2863049,
      "author_name": "coderrkj",
      "author_url": "",
      "post_date": "06/09/2024 07:40:43",
      "content": "<p>You should be able to train the models in any other enviroment (e.g., locally, Colab, cloud VM). But final submission must be done on a Kaggle Notebook with internet turned off. You can upload the trained model weights as a private dataset and load the weights during the model run.</p>\n<p>Me personally, I have not hit the limits of Kaggle for training as of yet, so I am continuing to use it. I am not certain what would be the best external option for this competition, but I would suggest looking into exhausting both GPU and TPU options in Kaggle before trying other alternatives.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2863812,
          "author_name": "devsya",
          "author_url": "",
          "post_date": "06/09/2024 17:43:08",
          "content": "<p>You should be able to train the models in any other environment (e.g., locally, Colab, cloud VM). But the final submission must be done on a Kaggle Notebook with the internet turned off. You can upload the trained model weights as a private dataset and load the weights during the model run.</p>\n<p><a href=\"https://www.kaggle.com/coderrkj\" target=\"_blank\">@coderrkj</a> I think this may be against the rules unless the pre-trained model is made publicly and freely available. 'Freely &amp; publicly available external data is allowed, including pre-trained models'- See the code requirements in the overview page. You can also think of it this way: if private pre-trained models were allowed wouldn't it be unfair to others with less compute resources?  Rules such as CPU Notebook &lt;= 9 hours run-time<br>\nGPU Notebook &lt;= 9 hours run-time would not make much sense if private pre-trained models were allowed, but I am not 100% sure</p>",
          "votes": null,
          "replies": [
            {
              "id": 2864493,
              "author_name": "mrnosaj",
              "author_url": "",
              "post_date": "06/10/2024 07:03:52",
              "content": "<p>hmmm good point</p>",
              "votes": null,
              "replies": []
            }
          ]
        },
        {
          "id": 2864492,
          "author_name": "mrnosaj",
          "author_url": "",
          "post_date": "06/10/2024 07:03:27",
          "content": "<p>Thanks for your advice it is greatly appreciated. I'll keep pursuing the Kaggle platform for now I think. </p>\n<p>The 'no internet' element was another learning curve for me. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2864161,
      "author_name": "devsya",
      "author_url": "",
      "post_date": "06/10/2024 01:48:58",
      "content": "<p>I have noticed that my code takes seconds to completely run but scoring takes about 9 to 15mins and when I used ResNet18 running took seconds but scoring took 2 hours. But I do not know why</p>",
      "votes": null,
      "replies": [
        {
          "id": 2864300,
          "author_name": "coderrkj",
          "author_url": "",
          "post_date": "06/10/2024 05:00:51",
          "content": "<p>The number of test cases are greater during submission. You need to account for the runtime of ResNet18 on those cases also in the 2 hours it takes during submission.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2863019": "Hi everyone,\n\nI've been participating in the RSNA 2024 Lumbar Spine Degenerative Classification competition and have been encountering frequent issues with exceeding the platform's limits. I was wondering how others are managing their code execution environments. Are you running your code exclusively on Kaggle, or are you using local machines or other platforms to develop and test your models before submitting?\n\nAny insights or recommendations would be greatly appreciated.\n\nThank you!",
    "2863049": "You should be able to train the models in any other enviroment (e.g., locally, Colab, cloud VM). But final submission must be done on a Kaggle Notebook with internet turned off. You can upload the trained model weights as a private dataset and load the weights during the model run.\n\nMe personally, I have not hit the limits of Kaggle for training as of yet, so I am continuing to use it. I am not certain what would be the best external option for this competition, but I would suggest looking into exhausting both GPU and TPU options in Kaggle before trying other alternatives.",
    "2863812": "You should be able to train the models in any other environment (e.g., locally, Colab, cloud VM). But the final submission must be done on a Kaggle Notebook with the internet turned off. You can upload the trained model weights as a private dataset and load the weights during the model run.\n\n@coderrkj I think this may be against the rules unless the pre-trained model is made publicly and freely available. 'Freely & publicly available external data is allowed, including pre-trained models'- See the code requirements in the overview page. You can also think of it this way: if private pre-trained models were allowed wouldn't it be unfair to others with less compute resources?  Rules such as CPU Notebook <= 9 hours run-time\nGPU Notebook <= 9 hours run-time would not make much sense if private pre-trained models were allowed, but I am not 100% sure",
    "2864161": "I have noticed that my code takes seconds to completely run but scoring takes about 9 to 15mins and when I used ResNet18 running took seconds but scoring took 2 hours. But I do not know why",
    "2864300": "The number of test cases are greater during submission. You need to account for the runtime of ResNet18 on those cases also in the 2 hours it takes during submission.",
    "2864492": "Thanks for your advice it is greatly appreciated. I'll keep pursuing the Kaggle platform for now I think. \n\nThe 'no internet' element was another learning curve for me.",
    "2864493": "hmmm good point"
  },
  "source": "meta"
}