{
  "id": 21380,
  "title": "Accuracy in relation to mean location",
  "url": "/competitions/facebook-v-predicting-check-ins/discussion/21380",
  "author_name": "",
  "post_date": "2016-06-02T08:44:17.147Z",
  "votes": 4,
  "comment_count": 2,
  "views": 846,
  "content": "<p>I still try to figure out how to incorporate the accuracy column and did some plotting.</p>\n\n<p>I calculated the average coordinate of every locationId in the train dataset and afterwards calculated the mean distance for every accuracy to the average locationId coordinate. I droped the accuracies with less than 10 occurances.</p>\n\n<p><img src=\"http://i.imgur.com/KO3ZqNy.png\" alt=\"enter image description here\" title>\nx-Axis: accuracy</p>\n\n<p>y-Axis: mean distance to location center</p>\n\n<p>This seems contrary to the popular opinion that the higher the accuracy the higher the confidence of the coordinate or do I miss something here?</p>",
  "messages": [
    {
      "id": "122238",
      "postDate": "06/02/2016 08:44:17",
      "content": "<p>I still try to figure out how to incorporate the accuracy column and did some plotting.</p>\n\n<p>I calculated the average coordinate of every locationId in the train dataset and afterwards calculated the mean distance for every accuracy to the average locationId coordinate. I droped the accuracies with less than 10 occurances.</p>\n\n<p><img src=\"http://i.imgur.com/KO3ZqNy.png\" alt=\"enter image description here\" title>\nx-Axis: accuracy</p>\n\n<p>y-Axis: mean distance to location center</p>\n\n<p>This seems contrary to the popular opinion that the higher the accuracy the higher the confidence of the coordinate or do I miss something here?</p>",
      "rawMarkdown": "I still try to figure out how to incorporate the accuracy column and did some plotting.\r\n\r\nI calculated the average coordinate of every locationId in the train dataset and afterwards calculated the mean distance for every accuracy to the average locationId coordinate. I droped the accuracies with less than 10 occurances.\r\n\r\n![enter image description here][1]\r\nx-Axis: accuracy\r\n\r\ny-Axis: mean distance to location center\r\n\r\nThis seems contrary to the popular opinion that the higher the accuracy the higher the confidence of the coordinate or do I miss something here?\r\n\r\n  [1]: http://i.imgur.com/KO3ZqNy.png",
      "votes": null
    },
    {
      "id": "122239",
      "postDate": "06/02/2016 08:46:13",
      "content": "<p>I added the raw csv of the data</p>\n\n<p>1 column: accuracy</p>\n\n<p>2 column: mean distance</p>\n\n<p>3 column: accuracy occurences</p>",
      "rawMarkdown": "I added the raw csv of the data\r\n\r\n1 column: accuracy\r\n\r\n2 column: mean distance\r\n\r\n3 column: accuracy occurences",
      "votes": null
    },
    {
      "id": "122945",
      "postDate": "06/08/2016 18:08:38",
      "content": "<p>I think the result might have something to do with how the mean distance is calculated. You gave the same weight to each accuracy. Since the accuracy peaks at around 70, the events with this accuracy would dominate in your mean calculation. This sort of explains why there's a significant dip in your plot around that value. I just get started so I can't be very sure about it, but I think if you assign each event different weight with respect to its accuracy you'll get something different.  </p>",
      "rawMarkdown": "I think the result might have something to do with how the mean distance is calculated. You gave the same weight to each accuracy. Since the accuracy peaks at around 70, the events with this accuracy would dominate in your mean calculation. This sort of explains why there's a significant dip in your plot around that value. I just get started so I can't be very sure about it, but I think if you assign each event different weight with respect to its accuracy you'll get something different.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 122239,
      "author_name": "scipioaemilianus",
      "author_url": "",
      "post_date": "06/02/2016 08:46:13",
      "content": "<p>I added the raw csv of the data</p>\n\n<p>1 column: accuracy</p>\n\n<p>2 column: mean distance</p>\n\n<p>3 column: accuracy occurences</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 122945,
      "author_name": "haothecabbage",
      "author_url": "",
      "post_date": "06/08/2016 18:08:38",
      "content": "<p>I think the result might have something to do with how the mean distance is calculated. You gave the same weight to each accuracy. Since the accuracy peaks at around 70, the events with this accuracy would dominate in your mean calculation. This sort of explains why there's a significant dip in your plot around that value. I just get started so I can't be very sure about it, but I think if you assign each event different weight with respect to its accuracy you'll get something different.  </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "122238": "I still try to figure out how to incorporate the accuracy column and did some plotting.\r\n\r\nI calculated the average coordinate of every locationId in the train dataset and afterwards calculated the mean distance for every accuracy to the average locationId coordinate. I droped the accuracies with less than 10 occurances.\r\n\r\n![enter image description here][1]\r\nx-Axis: accuracy\r\n\r\ny-Axis: mean distance to location center\r\n\r\nThis seems contrary to the popular opinion that the higher the accuracy the higher the confidence of the coordinate or do I miss something here?\r\n\r\n  [1]: http://i.imgur.com/KO3ZqNy.png",
    "122239": "I added the raw csv of the data\r\n\r\n1 column: accuracy\r\n\r\n2 column: mean distance\r\n\r\n3 column: accuracy occurences",
    "122945": "I think the result might have something to do with how the mean distance is calculated. You gave the same weight to each accuracy. Since the accuracy peaks at around 70, the events with this accuracy would dominate in your mean calculation. This sort of explains why there's a significant dip in your plot around that value. I just get started so I can't be very sure about it, but I think if you assign each event different weight with respect to its accuracy you'll get something different."
  },
  "source": "meta"
}