{
  "id": 27837,
  "title": "New to Kaggle: how should we approach images that are neither cats nor dogs in training data?",
  "url": "/competitions/dogs-vs-cats-redux-kernels-edition/discussion/27837",
  "author_name": "",
  "post_date": "2017-01-17T21:44:13.557Z",
  "votes": 1,
  "comment_count": 1,
  "views": 136,
  "content": "<p>I examined some of the incorrectly classified images in my validation set and there are e.g. an image with a red text \"Adopted\" on a white background, without cat or dog.</p>\n\n<p>How we should approach such images?\n Remove them from training data? \nMove them from validation data to training data?</p>",
  "messages": [
    {
      "id": "156787",
      "postDate": "01/17/2017 21:44:13",
      "content": "<p>I examined some of the incorrectly classified images in my validation set and there are e.g. an image with a red text \"Adopted\" on a white background, without cat or dog.</p>\n\n<p>How we should approach such images?\n Remove them from training data? \nMove them from validation data to training data?</p>",
      "rawMarkdown": "I examined some of the incorrectly classified images in my validation set and there are e.g. an image with a red text \"Adopted\" on a white background, without cat or dog.\r\n\r\nHow we should approach such images?\r\n Remove them from training data? \r\nMove them from validation data to training data?",
      "votes": null
    },
    {
      "id": "157639",
      "postDate": "01/22/2017 20:59:58",
      "content": "<p>I'm not an expert, but I would remove such images from the data. If a sample does not belong to one of the cases you are trying to predict, why keep it? It will only impair your model's ability to classify correctly. </p>",
      "rawMarkdown": "I'm not an expert, but I would remove such images from the data. If a sample does not belong to one of the cases you are trying to predict, why keep it? It will only impair your model's ability to classify correctly.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 157639,
      "author_name": "emptycasket",
      "author_url": "",
      "post_date": "01/22/2017 20:59:58",
      "content": "<p>I'm not an expert, but I would remove such images from the data. If a sample does not belong to one of the cases you are trying to predict, why keep it? It will only impair your model's ability to classify correctly. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "156787": "I examined some of the incorrectly classified images in my validation set and there are e.g. an image with a red text \"Adopted\" on a white background, without cat or dog.\r\n\r\nHow we should approach such images?\r\n Remove them from training data? \r\nMove them from validation data to training data?",
    "157639": "I'm not an expert, but I would remove such images from the data. If a sample does not belong to one of the cases you are trying to predict, why keep it? It will only impair your model's ability to classify correctly."
  },
  "source": "meta"
}