{
  "id": 5343,
  "title": "How were the questions generated?",
  "url": "/competitions/accelerometer-biometric-competition/discussion/5343",
  "author_name": "",
  "post_date": "2013-08-08T11:27:54.293Z",
  "votes": null,
  "comment_count": 7,
  "views": 3038,
  "content": "<p>To prepare a validation set it would be useful to know how were the questions generated. Particularly:</p>\n<ul>\n<li>what is the ratio of true positive/negative answers - 50/50?</li>\n<li>in case of negative answers, that is examples where a device id in the question is not the real device id, how were these labels chosen - randomly, or according to some other schema?</li>\n</ul>",
  "messages": [
    {
      "id": "28426",
      "postDate": "08/08/2013 11:27:54",
      "content": "<p>To prepare a validation set it would be useful to know how were the questions generated. Particularly:</p>\n<ul>\n<li>what is the ratio of true positive/negative answers - 50/50?</li>\n<li>in case of negative answers, that is examples where a device id in the question is not the real device id, how were these labels chosen - randomly, or according to some other schema?</li>\n</ul>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "28427",
      "postDate": "08/08/2013 12:04:34",
      "content": "<p>Yes ratio is 50/50</p>\n<p>Negative answers were randomly chosen from devices of the same make whenever possible; in cases of single device models in the data set, negatives were selected from devices with similar sampling frequencies.</p>\n<p>&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "29256",
      "postDate": "08/22/2013 19:01:38",
      "content": "<p>[quote=Geoff;28427]</p>\n<p>Yes ratio is 50/50</p>\n<p>Negative answers were randomly chosen from devices of the same make whenever possible; in cases of single device models in the data set, negatives were selected from devices with similar sampling frequencies.</p>\n<p>&nbsp;</p>\n<p>[/quote]</p>\n<p>&nbsp;</p>\n<p>Hi Geoff. Could you please explain what the positive negative answers is about? What I understand from the quests_dt file is that each device data is divided into subpieces each with 300 datapoints call sequenceID. We do the training on the train set where we only have the deviceID's. Then for testing, what we need to do is taking the sequenceID's in the test set, and assign each of them to a deviceID. The quest_dt file contains the correct answers. We need to compare our results with the quests_dt file.</p>\n<p>I might be wrong. Please correct me. Thank you.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "29262",
      "postDate": "08/22/2013 19:23:23",
      "content": "<p>Hey Ilyas, the train file has samples labeled with Device Id. The&nbsp;test file has data split into sequences of 300 samples each that are labeled with Sequence Id. The&nbsp;questions file matches each sequence to a Device Id. The goal is to correctly predict the probability that a stated match is&nbsp; correct (positive) or incorrect (negative).&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "29416",
      "postDate": "08/25/2013 01:38:17",
      "content": "<p>[quote=Geoff;28427]</p>\n<p><span style=\"line-height: 1.4\">in cases of single device models in the data set, negatives were selected from devices with similar sampling frequencies.</span>&nbsp;</p>\n<p>[/quote]</p>\n<p>Could you please elaborate on what you mean by sampling frequencies here? Is it how often a sample is collected, or does it refer to the number of samples collected from a device?</p>\n<p>&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "29421",
      "postDate": "08/25/2013 05:14:08",
      "content": "<p>The interval between consecutive samples.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "29513",
      "postDate": "08/26/2013 20:53:38",
      "content": "<p>Since the sampling frequency is not consistent, would you take the median, mean, or mode of the sampling frequencies? Or other.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "29578",
      "postDate": "08/27/2013 16:12:17",
      "content": "",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 28427,
      "author_name": "geoff3",
      "author_url": "",
      "post_date": "08/08/2013 12:04:34",
      "content": "<p>Yes ratio is 50/50</p>\n<p>Negative answers were randomly chosen from devices of the same make whenever possible; in cases of single device models in the data set, negatives were selected from devices with similar sampling frequencies.</p>\n<p>&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 29256,
      "author_name": "",
      "author_url": "",
      "post_date": "08/22/2013 19:01:38",
      "content": "<p>[quote=Geoff;28427]</p>\n<p>Yes ratio is 50/50</p>\n<p>Negative answers were randomly chosen from devices of the same make whenever possible; in cases of single device models in the data set, negatives were selected from devices with similar sampling frequencies.</p>\n<p>&nbsp;</p>\n<p>[/quote]</p>\n<p>&nbsp;</p>\n<p>Hi Geoff. Could you please explain what the positive negative answers is about? What I understand from the quests_dt file is that each device data is divided into subpieces each with 300 datapoints call sequenceID. We do the training on the train set where we only have the deviceID's. Then for testing, what we need to do is taking the sequenceID's in the test set, and assign each of them to a deviceID. The quest_dt file contains the correct answers. We need to compare our results with the quests_dt file.</p>\n<p>I might be wrong. Please correct me. Thank you.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 29262,
      "author_name": "geoff3",
      "author_url": "",
      "post_date": "08/22/2013 19:23:23",
      "content": "<p>Hey Ilyas, the train file has samples labeled with Device Id. The&nbsp;test file has data split into sequences of 300 samples each that are labeled with Sequence Id. The&nbsp;questions file matches each sequence to a Device Id. The goal is to correctly predict the probability that a stated match is&nbsp; correct (positive) or incorrect (negative).&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 29416,
      "author_name": "solorzano",
      "author_url": "",
      "post_date": "08/25/2013 01:38:17",
      "content": "<p>[quote=Geoff;28427]</p>\n<p><span style=\"line-height: 1.4\">in cases of single device models in the data set, negatives were selected from devices with similar sampling frequencies.</span>&nbsp;</p>\n<p>[/quote]</p>\n<p>Could you please elaborate on what you mean by sampling frequencies here? Is it how often a sample is collected, or does it refer to the number of samples collected from a device?</p>\n<p>&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 29421,
      "author_name": "geoff3",
      "author_url": "",
      "post_date": "08/25/2013 05:14:08",
      "content": "<p>The interval between consecutive samples.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 29513,
      "author_name": "garrettargenna",
      "author_url": "",
      "post_date": "08/26/2013 20:53:38",
      "content": "<p>Since the sampling frequency is not consistent, would you take the median, mean, or mode of the sampling frequencies? Or other.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 29578,
      "author_name": "cristiancanton",
      "author_url": "",
      "post_date": "08/27/2013 16:12:17",
      "content": "",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "28426": "",
    "28427": "",
    "29256": "",
    "29262": "",
    "29416": "",
    "29421": "",
    "29513": "",
    "29578": ""
  },
  "source": "meta"
}