{
  "id": 44458,
  "title": "How to use test data",
  "url": "/competitions/tensorflow-speech-recognition-challenge/discussion/44458",
  "author_name": "",
  "post_date": "2017-11-29T02:18:25.641183800Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi\nmaybe im missing something, but how do you use the test data if its not labelled?\ndo you have to manually check that your performance is good?</p>",
  "messages": [
    {
      "id": "249697",
      "postDate": "11/29/2017 02:18:25",
      "content": "<p>Hi\nmaybe im missing something, but how do you use the test data if its not labelled?\ndo you have to manually check that your performance is good?</p>",
      "rawMarkdown": "Hi\nmaybe im missing something, but how do you use the test data if its not labelled?\ndo you have to manually check that your performance is good?",
      "votes": null
    },
    {
      "id": "249844",
      "postDate": "11/29/2017 10:06:38",
      "content": "<p>If the test data was labelled, then everyone would score 100% on the competition. ;-) To check whether your performance is good you can make submissions to Kaggle and see what your position on the leaderboard is, but it's smarter to create a local test set from a portion of the training data. You don't train on this local test set, and only use it to confirm whether your model works well or not.</p>",
      "rawMarkdown": "If the test data was labelled, then everyone would score 100% on the competition. ;-) To check whether your performance is good you can make submissions to Kaggle and see what your position on the leaderboard is, but it's smarter to create a local test set from a portion of the training data. You don't train on this local test set, and only use it to confirm whether your model works well or not.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 249844,
      "author_name": "humananalog",
      "author_url": "",
      "post_date": "11/29/2017 10:06:38",
      "content": "<p>If the test data was labelled, then everyone would score 100% on the competition. ;-) To check whether your performance is good you can make submissions to Kaggle and see what your position on the leaderboard is, but it's smarter to create a local test set from a portion of the training data. You don't train on this local test set, and only use it to confirm whether your model works well or not.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "249697": "Hi\nmaybe im missing something, but how do you use the test data if its not labelled?\ndo you have to manually check that your performance is good?",
    "249844": "If the test data was labelled, then everyone would score 100% on the competition. ;-) To check whether your performance is good you can make submissions to Kaggle and see what your position on the leaderboard is, but it's smarter to create a local test set from a portion of the training data. You don't train on this local test set, and only use it to confirm whether your model works well or not."
  },
  "source": "meta"
}