{
  "id": 123522,
  "title": "Query about Kernel Only Part.",
  "url": "/competitions/deepfake-detection-challenge/discussion/123522",
  "author_name": "",
  "post_date": "2019-12-28T11:43:01.785439100Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>The full dataset is 471 GB. So we can probably cannot train a model on Kaggle server only right? </p>\n\n<p>So, I suppose we can train the model externally. </p>\n\n<p>But we need to import the weights of the model to Kaggle kernel, and load, process test data video on Kaggle kernel? So that users do not manually label the videos and submit. Only 400 videos, so it's possible and easy.</p>\n\n<p>Can we use external libraries such as Dlib? </p>",
  "messages": [
    {
      "id": "705085",
      "postDate": "12/28/2019 11:43:01",
      "content": "<p>The full dataset is 471 GB. So we can probably cannot train a model on Kaggle server only right? </p>\n\n<p>So, I suppose we can train the model externally. </p>\n\n<p>But we need to import the weights of the model to Kaggle kernel, and load, process test data video on Kaggle kernel? So that users do not manually label the videos and submit. Only 400 videos, so it's possible and easy.</p>\n\n<p>Can we use external libraries such as Dlib? </p>",
      "rawMarkdown": "The full dataset is 471 GB. So we can probably cannot train a model on Kaggle server only right? \n\nSo, I suppose we can train the model externally. \n\nBut we need to import the weights of the model to Kaggle kernel, and load, process test data video on Kaggle kernel? So that users do not manually label the videos and submit. Only 400 videos, so it's possible and easy.\n\nCan we use external libraries such as Dlib?",
      "votes": null
    },
    {
      "id": "705239",
      "postDate": "12/28/2019 17:19:53",
      "content": "<p>You cannot train the model on Kaggle at all, as the full dataset isn't available in the kernels.</p>\n\n<p>So you're correct that you need to import the weights of your model into the Kaggle kernel and then process the test videos to make predictions. However, there are 4000 -- not 400 -- test videos for the leaderboard submission.</p>",
      "rawMarkdown": "You cannot train the model on Kaggle at all, as the full dataset isn't available in the kernels.\n\nSo you're correct that you need to import the weights of your model into the Kaggle kernel and then process the test videos to make predictions. However, there are 4000 -- not 400 -- test videos for the leaderboard submission.",
      "votes": null
    },
    {
      "id": "706758",
      "postDate": "12/30/2019 19:38:45",
      "content": "<p>You may use external libraries, and would have to import it as an external data source, if it is not already in Kaggle's image.</p>",
      "rawMarkdown": "You may use external libraries, and would have to import it as an external data source, if it is not already in Kaggle's image.",
      "votes": null
    },
    {
      "id": "707275",
      "postDate": "12/31/2019 14:31:08",
      "content": "<p>How can I install an external library without using the Internet, as it's not allowed?</p>",
      "rawMarkdown": "How can I install an external library without using the Internet, as it's not allowed?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 705239,
      "author_name": "humananalog",
      "author_url": "",
      "post_date": "12/28/2019 17:19:53",
      "content": "<p>You cannot train the model on Kaggle at all, as the full dataset isn't available in the kernels.</p>\n\n<p>So you're correct that you need to import the weights of your model into the Kaggle kernel and then process the test videos to make predictions. However, there are 4000 -- not 400 -- test videos for the leaderboard submission.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 706758,
      "author_name": "juliaelliott",
      "author_url": "",
      "post_date": "12/30/2019 19:38:45",
      "content": "<p>You may use external libraries, and would have to import it as an external data source, if it is not already in Kaggle's image.</p>",
      "votes": null,
      "replies": [
        {
          "id": 707275,
          "author_name": "kamalchhirang",
          "author_url": "",
          "post_date": "12/31/2019 14:31:08",
          "content": "<p>How can I install an external library without using the Internet, as it's not allowed?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "705085": "The full dataset is 471 GB. So we can probably cannot train a model on Kaggle server only right? \n\nSo, I suppose we can train the model externally. \n\nBut we need to import the weights of the model to Kaggle kernel, and load, process test data video on Kaggle kernel? So that users do not manually label the videos and submit. Only 400 videos, so it's possible and easy.\n\nCan we use external libraries such as Dlib?",
    "705239": "You cannot train the model on Kaggle at all, as the full dataset isn't available in the kernels.\n\nSo you're correct that you need to import the weights of your model into the Kaggle kernel and then process the test videos to make predictions. However, there are 4000 -- not 400 -- test videos for the leaderboard submission.",
    "706758": "You may use external libraries, and would have to import it as an external data source, if it is not already in Kaggle's image.",
    "707275": "How can I install an external library without using the Internet, as it's not allowed?"
  },
  "source": "meta"
}