{
  "id": 21855,
  "title": "accuracy and data for periodic time that hit the boundary makes cv worse",
  "url": "/competitions/facebook-v-predicting-check-ins/discussion/21855",
  "author_name": "",
  "post_date": "2016-06-23T13:39:31.633Z",
  "votes": 1,
  "comment_count": 1,
  "views": 342,
  "content": "<p>observed that many are using accuracy feature (df['accuracy'] = np.log10(df['accuracy']) * 10.0) and  are adding data for periodic time that hit the boundary but then i try to do the same it definitively makes my cv score worse do you have the same inconsistency between cv score and lb score?</p>",
  "messages": [
    {
      "id": "124919",
      "postDate": "06/23/2016 13:39:31",
      "content": "<p>observed that many are using accuracy feature (df['accuracy'] = np.log10(df['accuracy']) * 10.0) and  are adding data for periodic time that hit the boundary but then i try to do the same it definitively makes my cv score worse do you have the same inconsistency between cv score and lb score?</p>",
      "rawMarkdown": "observed that many are using accuracy feature (df['accuracy'] = np.log10(df['accuracy']) * 10.0) and  are adding data for periodic time that hit the boundary but then i try to do the same it definitively makes my cv score worse do you have the same inconsistency between cv score and lb score?",
      "votes": null
    },
    {
      "id": "125011",
      "postDate": "06/24/2016 16:00:04",
      "content": "<p>I tested them on my local validation set. You are right, the accuracy scaling and adding data for periodic time DO make my local score worse. I didn't try it on LB.</p>",
      "rawMarkdown": "I tested them on my local validation set. You are right, the accuracy scaling and adding data for periodic time DO make my local score worse. I didn't try it on LB.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 125011,
      "author_name": "aikinogard",
      "author_url": "",
      "post_date": "06/24/2016 16:00:04",
      "content": "<p>I tested them on my local validation set. You are right, the accuracy scaling and adding data for periodic time DO make my local score worse. I didn't try it on LB.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "124919": "observed that many are using accuracy feature (df['accuracy'] = np.log10(df['accuracy']) * 10.0) and  are adding data for periodic time that hit the boundary but then i try to do the same it definitively makes my cv score worse do you have the same inconsistency between cv score and lb score?",
    "125011": "I tested them on my local validation set. You are right, the accuracy scaling and adding data for periodic time DO make my local score worse. I didn't try it on LB."
  },
  "source": "meta"
}