{
  "id": 230618,
  "title": "Clarification on compute time",
  "url": "/competitions/coleridgeinitiative-show-us-the-data/discussion/230618",
  "author_name": "",
  "post_date": "2021-04-04T21:14:55.138021100Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>The rules state that the code has to run inside a notebook within the allotted time in order to be valid. Does this mean that any model training needs to run within that timeframe?</p>\n<p>Is uploading a model as a public dataset a workaround for this? This has the disadvantage that the model file is public for anyone to use. But then you could upload a public dataset with a model that is encrypted and only store the key inside your private submission notebook. This would obviously go against the spirit of a public dataset, but it seems like a way to cheat around the limitations of the competition. ( I'm not planning to do this! Just trying to understand if this is an issue or not. 😄)</p>\n<p>Sorry for the noob question, I haven't participated in any competitions before this.</p>",
  "messages": [
    {
      "id": "1262905",
      "postDate": "04/04/2021 21:14:55",
      "content": "<p>The rules state that the code has to run inside a notebook within the allotted time in order to be valid. Does this mean that any model training needs to run within that timeframe?</p>\n<p>Is uploading a model as a public dataset a workaround for this? This has the disadvantage that the model file is public for anyone to use. But then you could upload a public dataset with a model that is encrypted and only store the key inside your private submission notebook. This would obviously go against the spirit of a public dataset, but it seems like a way to cheat around the limitations of the competition. ( I'm not planning to do this! Just trying to understand if this is an issue or not. 😄)</p>\n<p>Sorry for the noob question, I haven't participated in any competitions before this.</p>",
      "rawMarkdown": "The rules state that the code has to run inside a notebook within the allotted time in order to be valid. Does this mean that any model training needs to run within that timeframe?\n\nIs uploading a model as a public dataset a workaround for this? This has the disadvantage that the model file is public for anyone to use. But then you could upload a public dataset with a model that is encrypted and only store the key inside your private submission notebook. This would obviously go against the spirit of a public dataset, but it seems like a way to cheat around the limitations of the competition. ( I'm not planning to do this! Just trying to understand if this is an issue or not. 😄)\n\nSorry for the noob question, I haven't participated in any competitions before this.",
      "votes": null
    },
    {
      "id": "1266114",
      "postDate": "04/07/2021 13:48:47",
      "content": "<p><a href=\"https://www.kaggle.com/philculliton\" target=\"_blank\">@philculliton</a> , would you be able to clarify wheter all model training needs to occur within the compute time limit?</p>\n<p>A somewhat related question: can we upload python packages which are not part of the default notebook environment as \"external datasets\"? What about compiled libraries derived from appropriately-licensed open source code?</p>",
      "rawMarkdown": "philculliton , would you be able to clarify wheter all model training needs to occur within the compute time limit?\n\nA somewhat related question: can we upload python packages which are not part of the default notebook environment as \"external datasets\"? What about compiled libraries derived from appropriately-licensed open source code?",
      "votes": null
    },
    {
      "id": "1266130",
      "postDate": "04/07/2021 14:05:34",
      "content": "<p><a href=\"https://www.kaggle.com/mskoh52\" target=\"_blank\">@mskoh52</a> You can train in another notebook/on your local machine/on a cloud and use the trained model for inference/submission (=you can use your model form a private notebook/dataset).</p>",
      "rawMarkdown": "mskoh52 You can train in another notebook/on your local machine/on a cloud and use the trained model for inference/submission (=you can use your model form a private notebook/dataset).",
      "votes": null
    },
    {
      "id": "1267325",
      "postDate": "04/08/2021 13:20:34",
      "content": "<blockquote>\n  <p><a href=\"https://www.kaggle.com/mskoh52\" target=\"_blank\">@mskoh52</a> You can train in another notebook/on your local machine/on a cloud and use the trained model for inference/submission (=you can use your model form a private notebook/dataset).</p>\n</blockquote>\n<p>Good to know, thank you!</p>",
      "rawMarkdown": "> @mskoh52 You can train in another notebook/on your local machine/on a cloud and use the trained model for inference/submission (=you can use your model form a private notebook/dataset).\n\nGood to know, thank you!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1266114,
      "author_name": "mskoh52",
      "author_url": "",
      "post_date": "04/07/2021 13:48:47",
      "content": "<p><a href=\"https://www.kaggle.com/philculliton\" target=\"_blank\">@philculliton</a> , would you be able to clarify wheter all model training needs to occur within the compute time limit?</p>\n<p>A somewhat related question: can we upload python packages which are not part of the default notebook environment as \"external datasets\"? What about compiled libraries derived from appropriately-licensed open source code?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1266130,
      "author_name": "hannes82",
      "author_url": "",
      "post_date": "04/07/2021 14:05:34",
      "content": "<p><a href=\"https://www.kaggle.com/mskoh52\" target=\"_blank\">@mskoh52</a> You can train in another notebook/on your local machine/on a cloud and use the trained model for inference/submission (=you can use your model form a private notebook/dataset).</p>",
      "votes": null,
      "replies": [
        {
          "id": 1267325,
          "author_name": "mskoh52",
          "author_url": "",
          "post_date": "04/08/2021 13:20:34",
          "content": "<blockquote>\n  <p><a href=\"https://www.kaggle.com/mskoh52\" target=\"_blank\">@mskoh52</a> You can train in another notebook/on your local machine/on a cloud and use the trained model for inference/submission (=you can use your model form a private notebook/dataset).</p>\n</blockquote>\n<p>Good to know, thank you!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1262905": "The rules state that the code has to run inside a notebook within the allotted time in order to be valid. Does this mean that any model training needs to run within that timeframe?\n\nIs uploading a model as a public dataset a workaround for this? This has the disadvantage that the model file is public for anyone to use. But then you could upload a public dataset with a model that is encrypted and only store the key inside your private submission notebook. This would obviously go against the spirit of a public dataset, but it seems like a way to cheat around the limitations of the competition. ( I'm not planning to do this! Just trying to understand if this is an issue or not. 😄)\n\nSorry for the noob question, I haven't participated in any competitions before this.",
    "1266114": "philculliton , would you be able to clarify wheter all model training needs to occur within the compute time limit?\n\nA somewhat related question: can we upload python packages which are not part of the default notebook environment as \"external datasets\"? What about compiled libraries derived from appropriately-licensed open source code?",
    "1266130": "mskoh52 You can train in another notebook/on your local machine/on a cloud and use the trained model for inference/submission (=you can use your model form a private notebook/dataset).",
    "1267325": "> @mskoh52 You can train in another notebook/on your local machine/on a cloud and use the trained model for inference/submission (=you can use your model form a private notebook/dataset).\n\nGood to know, thank you!"
  },
  "source": "meta"
}