{
  "id": 14212,
  "title": "Some Humans wrongly classified as Bots in training set?",
  "url": "/competitions/facebook-recruiting-iv-human-or-bot/discussion/14212",
  "author_name": "",
  "post_date": "2015-05-14T02:05:15.207Z",
  "votes": 6,
  "comment_count": 1,
  "views": 1116,
  "content": "<p>As some people have pointed out in other threads, there seem to be some bidders with non-human like characteristics that have still been classified as humans.&nbsp;</p>\n<p>If that's the case, it maybe a good idea to either correct the labels or disregard these anamolies for training the model?</p>\n\n<p>But, how can we be sure that we won't have mis-classified observations in the test set as well? In other words, if the ground truth is on shaky ground, isn't the judging criteria itself flawed?</p>\n<p>Can the admins vouch for the fact that all labels have been correctly assigned in both the training and test data?</p>",
  "messages": [
    {
      "id": "78346",
      "postDate": "05/14/2015 02:05:15",
      "content": "<p>As some people have pointed out in other threads, there seem to be some bidders with non-human like characteristics that have still been classified as humans.&nbsp;</p>\n<p>If that's the case, it maybe a good idea to either correct the labels or disregard these anamolies for training the model?</p>\n\n<p>But, how can we be sure that we won't have mis-classified observations in the test set as well? In other words, if the ground truth is on shaky ground, isn't the judging criteria itself flawed?</p>\n<p>Can the admins vouch for the fact that all labels have been correctly assigned in both the training and test data?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78452",
      "postDate": "05/15/2015 01:00:46",
      "content": "<p>There's some truth to&nbsp;this. If the algorithm that originally generated the labels is suboptimal, than we are actually competing to best approximate the suboptimal model, with whatever shortfalls it has. Though in practice, for a problem like this, I'm not sure how it would be avoided without the data being synthetic.&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 78452,
      "author_name": "ccphillippi",
      "author_url": "",
      "post_date": "05/15/2015 01:00:46",
      "content": "<p>There's some truth to&nbsp;this. If the algorithm that originally generated the labels is suboptimal, than we are actually competing to best approximate the suboptimal model, with whatever shortfalls it has. Though in practice, for a problem like this, I'm not sure how it would be avoided without the data being synthetic.&nbsp;</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "78346": "",
    "78452": ""
  },
  "source": "meta"
}