{
  "id": 131708,
  "title": "uploading output file on kaggle kernel for submission",
  "url": "/competitions/deepfake-detection-challenge/discussion/131708",
  "author_name": "",
  "post_date": "2020-02-21T08:09:30.762998700Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I generated an output csv file containing predictions of the results for the (public) test video from my trained model outside of the Kaggle kernel. Then I uploaded the output csv file on the Kaggle kernel, committed, then submitted the notebook but this resulted in a \"Submission Scoring Error\". Is it not possible just to upload a csv file of predictions (of the 400 test videos) on kernel just to see how it does on the public test data? or is the public test score calculated from more videos than the provided 400 public test videos? </p>",
  "messages": [
    {
      "id": "752616",
      "postDate": "02/21/2020 08:09:30",
      "content": "<p>I generated an output csv file containing predictions of the results for the (public) test video from my trained model outside of the Kaggle kernel. Then I uploaded the output csv file on the Kaggle kernel, committed, then submitted the notebook but this resulted in a \"Submission Scoring Error\". Is it not possible just to upload a csv file of predictions (of the 400 test videos) on kernel just to see how it does on the public test data? or is the public test score calculated from more videos than the provided 400 public test videos? </p>",
      "rawMarkdown": "I generated an output csv file containing predictions of the results for the (public) test video from my trained model outside of the Kaggle kernel. Then I uploaded the output csv file on the Kaggle kernel, committed, then submitted the notebook but this resulted in a \"Submission Scoring Error\". Is it not possible just to upload a csv file of predictions (of the 400 test videos) on kernel just to see how it does on the public test data? or is the public test score calculated from more videos than the provided 400 public test videos?",
      "votes": null
    },
    {
      "id": "752697",
      "postDate": "02/21/2020 10:08:20",
      "content": "<p>Yes, the 400 validation videos get changed to the 4000 'public test set' videos when you click submit, so providing predictions for those 400 videos isn't appropriate. For this reason, your code also needs to finish in ~54 minutes or less, given it has to process a dataset 10 times larger on submission within 9 hours.</p>",
      "rawMarkdown": "Yes, the 400 validation videos get changed to the 4000 'public test set' videos when you click submit, so providing predictions for those 400 videos isn't appropriate. For this reason, your code also needs to finish in ~54 minutes or less, given it has to process a dataset 10 times larger on submission within 9 hours.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 752697,
      "author_name": "jamesphoward",
      "author_url": "",
      "post_date": "02/21/2020 10:08:20",
      "content": "<p>Yes, the 400 validation videos get changed to the 4000 'public test set' videos when you click submit, so providing predictions for those 400 videos isn't appropriate. For this reason, your code also needs to finish in ~54 minutes or less, given it has to process a dataset 10 times larger on submission within 9 hours.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "752616": "I generated an output csv file containing predictions of the results for the (public) test video from my trained model outside of the Kaggle kernel. Then I uploaded the output csv file on the Kaggle kernel, committed, then submitted the notebook but this resulted in a \"Submission Scoring Error\". Is it not possible just to upload a csv file of predictions (of the 400 test videos) on kernel just to see how it does on the public test data? or is the public test score calculated from more videos than the provided 400 public test videos?",
    "752697": "Yes, the 400 validation videos get changed to the 4000 'public test set' videos when you click submit, so providing predictions for those 400 videos isn't appropriate. For this reason, your code also needs to finish in ~54 minutes or less, given it has to process a dataset 10 times larger on submission within 9 hours."
  },
  "source": "meta"
}