{
  "id": 23692,
  "title": "31 Features missing but measured in Date set",
  "url": "/competitions/bosch-production-line-performance/discussion/23692",
  "author_name": "",
  "post_date": "2016-09-15T20:42:39.207Z",
  "votes": 2,
  "comment_count": 3,
  "views": 764,
  "content": "<p>Hi hi dear data scientist fellows,</p>\n\n<p>While I was checking all the features that have been measured in date.csv, I noticed there are 31 features that been measured but does not exist in either numeric.csv or categorical.csv. </p>\n\n<p>I did shift the feature number by 1 as explained in data page. (e.g. L0_S0_D1 in date.csv is linked to L0_S0_F0 in numeric.csv)</p>\n\n<p>Am I missing something?  Any help or hint is appreciated!</p>\n\n<p>Missing features:\n[139, 140, 147, 456, 457, 464, 1825, 1832, 2234, 2235, 2444, 2445, 2452, 2501, 2502, 2727, 2728, 2735, 2784, 2785, 3010, 3011, 3018, 3874, 3942, 4009, 4012, 4021, 4024, 4194, 4254]</p>",
  "messages": [
    {
      "id": "135671",
      "postDate": "09/15/2016 20:42:39",
      "content": "<p>Hi hi dear data scientist fellows,</p>\n\n<p>While I was checking all the features that have been measured in date.csv, I noticed there are 31 features that been measured but does not exist in either numeric.csv or categorical.csv. </p>\n\n<p>I did shift the feature number by 1 as explained in data page. (e.g. L0_S0_D1 in date.csv is linked to L0_S0_F0 in numeric.csv)</p>\n\n<p>Am I missing something?  Any help or hint is appreciated!</p>\n\n<p>Missing features:\n[139, 140, 147, 456, 457, 464, 1825, 1832, 2234, 2235, 2444, 2445, 2452, 2501, 2502, 2727, 2728, 2735, 2784, 2785, 3010, 3011, 3018, 3874, 3942, 4009, 4012, 4021, 4024, 4194, 4254]</p>",
      "rawMarkdown": "Hi hi dear data scientist fellows,\r\n\r\nWhile I was checking all the features that have been measured in date.csv, I noticed there are 31 features that been measured but does not exist in either numeric.csv or categorical.csv. \r\n\r\nI did shift the feature number by 1 as explained in data page. (e.g. L0_S0_D1 in date.csv is linked to L0_S0_F0 in numeric.csv)\r\n\r\nAm I missing something?  Any help or hint is appreciated!\r\n\r\nMissing features:\r\n[139, 140, 147, 456, 457, 464, 1825, 1832, 2234, 2235, 2444, 2445, 2452, 2501, 2502, 2727, 2728, 2735, 2784, 2785, 3010, 3011, 3018, 3874, 3942, 4009, 4012, 4021, 4024, 4194, 4254]",
      "votes": null
    },
    {
      "id": "135673",
      "postDate": "09/15/2016 20:59:54",
      "content": "<p>In addition, not all features in num.csv and categorical.csv were included in date.csv.</p>\n\n<p>Only 366 numeric features and 759 categorical features have corresponding timestamps.</p>\n\n<p>My thought is, Bosch might have purposely removed many features in date.csv. I wonder why would they do this. (Potential info leakage? Or just to avoid providing too much of details of their business?)</p>",
      "rawMarkdown": "In addition, not all features in num.csv and categorical.csv were included in date.csv.\r\n\r\nOnly 366 numeric features and 759 categorical features have corresponding timestamps.\r\n\r\nMy thought is, Bosch might have purposely removed many features in date.csv. I wonder why would they do this. (Potential info leakage? Or just to avoid providing too much of details of their business?)",
      "votes": null
    },
    {
      "id": "135682",
      "postDate": "09/15/2016 22:38:00",
      "content": "<p>There has been a fair bit of discussion about the relationship between (train|test)_(numeric|categorical).csv.zip and the dates. See this <a href=\"https://www.kaggle.com/c/bosch-production-line-performance/forums/t/23037/l0-s0-d1-l0-s0-f0/132888#post132888\">earlier thread</a>, for a sample.\nClearly, there isn't a 1:1 correspondence between timestamps and features. My own guess is that the timestamps pertain to all of the trailing features back to the preceding timestamp. That is, some collection of k features n...n+k are collected at the same time and the timestamp n+k+1 pertains to all of them. But, of course, that's speculation on my part.     </p>\n\n<p>Organizers usually don't give out information after the start of the competition, so speculation may be all we are going to get.</p>",
      "rawMarkdown": "There has been a fair bit of discussion about the relationship between (train|test)_(numeric|categorical).csv.zip and the dates. See this [earlier thread][1], for a sample.\r\nClearly, there isn't a 1:1 correspondence between timestamps and features. My own guess is that the timestamps pertain to all of the trailing features back to the preceding timestamp. That is, some collection of k features n...n+k are collected at the same time and the timestamp n+k+1 pertains to all of them. But, of course, that's speculation on my part.     \r\n\r\nOrganizers usually don't give out information after the start of the competition, so speculation may be all we are going to get.\r\n\r\n  [1]: https://www.kaggle.com/c/bosch-production-line-performance/forums/t/23037/l0-s0-d1-l0-s0-f0/132888#post132888",
      "votes": null
    },
    {
      "id": "135685",
      "postDate": "09/16/2016 00:59:09",
      "content": "<p>Thank you, David!</p>",
      "rawMarkdown": "Thank you, David!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 135673,
      "author_name": "jonathanlxy",
      "author_url": "",
      "post_date": "09/15/2016 20:59:54",
      "content": "<p>In addition, not all features in num.csv and categorical.csv were included in date.csv.</p>\n\n<p>Only 366 numeric features and 759 categorical features have corresponding timestamps.</p>\n\n<p>My thought is, Bosch might have purposely removed many features in date.csv. I wonder why would they do this. (Potential info leakage? Or just to avoid providing too much of details of their business?)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 135682,
      "author_name": "davidthaler",
      "author_url": "",
      "post_date": "09/15/2016 22:38:00",
      "content": "<p>There has been a fair bit of discussion about the relationship between (train|test)_(numeric|categorical).csv.zip and the dates. See this <a href=\"https://www.kaggle.com/c/bosch-production-line-performance/forums/t/23037/l0-s0-d1-l0-s0-f0/132888#post132888\">earlier thread</a>, for a sample.\nClearly, there isn't a 1:1 correspondence between timestamps and features. My own guess is that the timestamps pertain to all of the trailing features back to the preceding timestamp. That is, some collection of k features n...n+k are collected at the same time and the timestamp n+k+1 pertains to all of them. But, of course, that's speculation on my part.     </p>\n\n<p>Organizers usually don't give out information after the start of the competition, so speculation may be all we are going to get.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 135685,
      "author_name": "jonathanlxy",
      "author_url": "",
      "post_date": "09/16/2016 00:59:09",
      "content": "<p>Thank you, David!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "135671": "Hi hi dear data scientist fellows,\r\n\r\nWhile I was checking all the features that have been measured in date.csv, I noticed there are 31 features that been measured but does not exist in either numeric.csv or categorical.csv. \r\n\r\nI did shift the feature number by 1 as explained in data page. (e.g. L0_S0_D1 in date.csv is linked to L0_S0_F0 in numeric.csv)\r\n\r\nAm I missing something?  Any help or hint is appreciated!\r\n\r\nMissing features:\r\n[139, 140, 147, 456, 457, 464, 1825, 1832, 2234, 2235, 2444, 2445, 2452, 2501, 2502, 2727, 2728, 2735, 2784, 2785, 3010, 3011, 3018, 3874, 3942, 4009, 4012, 4021, 4024, 4194, 4254]",
    "135673": "In addition, not all features in num.csv and categorical.csv were included in date.csv.\r\n\r\nOnly 366 numeric features and 759 categorical features have corresponding timestamps.\r\n\r\nMy thought is, Bosch might have purposely removed many features in date.csv. I wonder why would they do this. (Potential info leakage? Or just to avoid providing too much of details of their business?)",
    "135682": "There has been a fair bit of discussion about the relationship between (train|test)_(numeric|categorical).csv.zip and the dates. See this [earlier thread][1], for a sample.\r\nClearly, there isn't a 1:1 correspondence between timestamps and features. My own guess is that the timestamps pertain to all of the trailing features back to the preceding timestamp. That is, some collection of k features n...n+k are collected at the same time and the timestamp n+k+1 pertains to all of them. But, of course, that's speculation on my part.     \r\n\r\nOrganizers usually don't give out information after the start of the competition, so speculation may be all we are going to get.\r\n\r\n  [1]: https://www.kaggle.com/c/bosch-production-line-performance/forums/t/23037/l0-s0-d1-l0-s0-f0/132888#post132888",
    "135685": "Thank you, David!"
  },
  "source": "meta"
}