{
  "id": 22314,
  "title": "Is driver_imgs_list.csv useful?",
  "url": "/competitions/state-farm-distracted-driver-detection/discussion/22314",
  "author_name": "",
  "post_date": "2016-07-16T14:34:22.627Z",
  "votes": null,
  "comment_count": 1,
  "views": 396,
  "content": "<p>I checked train/test images and driver_imgs_list.csv, since train and test have different drivers, i think driver_imgs_list.csv file is useless, is that right?</p>",
  "messages": [
    {
      "id": "127950",
      "postDate": "07/16/2016 14:34:22",
      "content": "<p>I checked train/test images and driver_imgs_list.csv, since train and test have different drivers, i think driver_imgs_list.csv file is useless, is that right?</p>",
      "rawMarkdown": "I checked train/test images and driver_imgs_list.csv, since train and test have different drivers, i think driver_imgs_list.csv file is useless, is that right?",
      "votes": null
    },
    {
      "id": "127969",
      "postDate": "07/16/2016 16:25:26",
      "content": "<p>It is very useful in setting up local CV properly -- i.e. you can train on a subset of the unique drivers available (say 22), and validate it on the remaining 4 drivers. To generalize better, you might want to do several folds, each with a different set of drivers to train (and probably different augmentation like rotations and shifts).</p>",
      "rawMarkdown": "It is very useful in setting up local CV properly -- i.e. you can train on a subset of the unique drivers available (say 22), and validate it on the remaining 4 drivers. To generalize better, you might want to do several folds, each with a different set of drivers to train (and probably different augmentation like rotations and shifts).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 127969,
      "author_name": "yilihome",
      "author_url": "",
      "post_date": "07/16/2016 16:25:26",
      "content": "<p>It is very useful in setting up local CV properly -- i.e. you can train on a subset of the unique drivers available (say 22), and validate it on the remaining 4 drivers. To generalize better, you might want to do several folds, each with a different set of drivers to train (and probably different augmentation like rotations and shifts).</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "127950": "I checked train/test images and driver_imgs_list.csv, since train and test have different drivers, i think driver_imgs_list.csv file is useless, is that right?",
    "127969": "It is very useful in setting up local CV properly -- i.e. you can train on a subset of the unique drivers available (say 22), and validate it on the remaining 4 drivers. To generalize better, you might want to do several folds, each with a different set of drivers to train (and probably different augmentation like rotations and shifts)."
  },
  "source": "meta"
}