{
  "id": 112930,
  "title": "TIMESTAMP QUESTION",
  "url": "/competitions/ashrae-energy-prediction/discussion/112930",
  "author_name": "",
  "post_date": "2019-10-16T07:07:37.699800400Z",
  "votes": 4,
  "comment_count": 7,
  "views": 0,
  "content": "<p>So we have timestamp where we can extract information such as hour, month, year, weekday etc..</p>\n\n<p>But does it make sense to extract things such as weekday or year?</p>\n\n<p>I mean we know that winter is cold so month can tell us if its winter or not.\nWe know that nights are colder than midday so hour can also have some pattern in energy consumption.</p>\n\n<p>But I have seen some notebooks with weekday and year as features and to be honest.. Does it really make any sense to include those?</p>",
  "messages": [
    {
      "id": "650186",
      "postDate": "10/16/2019 07:07:37",
      "content": "<p>So we have timestamp where we can extract information such as hour, month, year, weekday etc..</p>\n\n<p>But does it make sense to extract things such as weekday or year?</p>\n\n<p>I mean we know that winter is cold so month can tell us if its winter or not.\nWe know that nights are colder than midday so hour can also have some pattern in energy consumption.</p>\n\n<p>But I have seen some notebooks with weekday and year as features and to be honest.. Does it really make any sense to include those?</p>",
      "rawMarkdown": "So we have timestamp where we can extract information such as hour, month, year, weekday etc..\n\nBut does it make sense to extract things such as weekday or year?\n\nI mean we know that winter is cold so month can tell us if its winter or not.\nWe know that nights are colder than midday so hour can also have some pattern in energy consumption.\n\nBut I have seen some notebooks with weekday and year as features and to be honest.. Does it really make any sense to include those?",
      "votes": null
    },
    {
      "id": "650311",
      "postDate": "10/16/2019 09:03:45",
      "content": "<blockquote>\n  <p>But I have seen some notebooks with weekday and year as features and to be honest</p>\n</blockquote>\n\n<p>Those may help. </p>\n\n<ul>\n<li><p>There are educational institutes/offices which may only operate during the weekdays (more energy consumption) and closed over the weekends (lesser energy consumption). Shopping malls may be more crowded over the weekends (Heavy energy consumption) and less crowded during the weekdays. Thus is_weekend is an important parameter. Similarly, holidays and days of yearly sales could be important features as well.</p></li>\n<li><p>The energy consumption of a building could be increasing year over year. But, since our training data belongs to just one year, not sure how effective it will be. May be we should build a feature with number of months since 2015 Jan (or something like that to capture the long term trend in the data)</p></li>\n</ul>\n\n<p>Finally, one thing which I have learned about ML from Jeremy Haward (from his fast.ai course) is : </p>\n\n<p>Question : \"Will blah help as a feature?\"\nAnswer: \"Try blah &amp; blah and see...\"</p>\n\n<p>Hope this helps.</p>",
      "rawMarkdown": "&gt; But I have seen some notebooks with weekday and year as features and to be honest\n\nThose may help. \n\n- There are educational institutes/offices which may only operate during the weekdays (more energy consumption) and closed over the weekends (lesser energy consumption). Shopping malls may be more crowded over the weekends (Heavy energy consumption) and less crowded during the weekdays. Thus is_weekend is an important parameter. Similarly, holidays and days of yearly sales could be important features as well.\n\n- The energy consumption of a building could be increasing year over year. But, since our training data belongs to just one year, not sure how effective it will be. May be we should build a feature with number of months since 2015 Jan (or something like that to capture the long term trend in the data)\n\nFinally, one thing which I have learned about ML from Jeremy Haward (from his fast.ai course) is : \n\nQuestion : \"Will blah help as a feature?\"\nAnswer: \"Try blah &amp; blah and see...\"\n\nHope this helps.",
      "votes": null
    },
    {
      "id": "650395",
      "postDate": "10/16/2019 10:18:54",
      "content": "<p>Thanks, that clears things out.</p>",
      "rawMarkdown": "Thanks, that clears things out.",
      "votes": null
    },
    {
      "id": "650548",
      "postDate": "10/16/2019 13:52:18",
      "content": "<p>Thanks arnab, for the explanations you provided. In fact, all of the above explanations are valid, since Energy is a resource that we all use, some more, some less, varying according to the days within the year. Thanks.</p>",
      "rawMarkdown": "Thanks arnab, for the explanations you provided. In fact, all of the above explanations are valid, since Energy is a resource that we all use, some more, some less, varying according to the days within the year. Thanks.",
      "votes": null
    },
    {
      "id": "650564",
      "postDate": "10/16/2019 14:06:38",
      "content": "<p>Thanks!</p>",
      "rawMarkdown": "Thanks!",
      "votes": null
    },
    {
      "id": "650863",
      "postDate": "10/16/2019 19:11:11",
      "content": "<p>Yes it does. buildings metadata stores building type, so you can extract \"working_hours\" factor with lower energy demand</p>",
      "rawMarkdown": "Yes it does. buildings metadata stores building type, so you can extract \"working_hours\" factor with lower energy demand",
      "votes": null
    },
    {
      "id": "651730",
      "postDate": "10/17/2019 21:28:04",
      "content": "<p>Maybe this is a dumb question but are these timestamp in the same time zone?</p>",
      "rawMarkdown": "Maybe this is a dumb question but are these timestamp in the same time zone?",
      "votes": null
    },
    {
      "id": "652703",
      "postDate": "10/19/2019 09:03:00",
      "content": "<p>I don't think timezone would matter anyway as midday in for example France carries the same characteristics as midday somewhere else. I think location matters more than timezone.</p>\n<p>Besides I don't think they run experiment on sites that far from each other, and also I think algorithms would be able to handle it as each site have unique site id.</p>\n<p>Correct me if I'm wrong.</p>",
      "rawMarkdown": "I don't think timezone would matter anyway as midday in for example France carries the same characteristics as midday somewhere else. I think location matters more than timezone.\n\nBesides I don't think they run experiment on sites that far from each other, and also I think algorithms would be able to handle it as each site have unique site id.\n\nCorrect me if I'm wrong.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 650311,
      "author_name": "arnabbiswas1",
      "author_url": "",
      "post_date": "10/16/2019 09:03:45",
      "content": "<blockquote>\n  <p>But I have seen some notebooks with weekday and year as features and to be honest</p>\n</blockquote>\n\n<p>Those may help. </p>\n\n<ul>\n<li><p>There are educational institutes/offices which may only operate during the weekdays (more energy consumption) and closed over the weekends (lesser energy consumption). Shopping malls may be more crowded over the weekends (Heavy energy consumption) and less crowded during the weekdays. Thus is_weekend is an important parameter. Similarly, holidays and days of yearly sales could be important features as well.</p></li>\n<li><p>The energy consumption of a building could be increasing year over year. But, since our training data belongs to just one year, not sure how effective it will be. May be we should build a feature with number of months since 2015 Jan (or something like that to capture the long term trend in the data)</p></li>\n</ul>\n\n<p>Finally, one thing which I have learned about ML from Jeremy Haward (from his fast.ai course) is : </p>\n\n<p>Question : \"Will blah help as a feature?\"\nAnswer: \"Try blah &amp; blah and see...\"</p>\n\n<p>Hope this helps.</p>",
      "votes": null,
      "replies": [
        {
          "id": 650395,
          "author_name": "alexkaggle95",
          "author_url": "",
          "post_date": "10/16/2019 10:18:54",
          "content": "<p>Thanks, that clears things out.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 650548,
      "author_name": "josebruno",
      "author_url": "",
      "post_date": "10/16/2019 13:52:18",
      "content": "<p>Thanks arnab, for the explanations you provided. In fact, all of the above explanations are valid, since Energy is a resource that we all use, some more, some less, varying according to the days within the year. Thanks.</p>",
      "votes": null,
      "replies": [
        {
          "id": 650564,
          "author_name": "arnabbiswas1",
          "author_url": "",
          "post_date": "10/16/2019 14:06:38",
          "content": "<p>Thanks!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 650863,
      "author_name": "michalpilch",
      "author_url": "",
      "post_date": "10/16/2019 19:11:11",
      "content": "<p>Yes it does. buildings metadata stores building type, so you can extract \"working_hours\" factor with lower energy demand</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 651730,
      "author_name": "terryduan",
      "author_url": "",
      "post_date": "10/17/2019 21:28:04",
      "content": "<p>Maybe this is a dumb question but are these timestamp in the same time zone?</p>",
      "votes": null,
      "replies": [
        {
          "id": 652703,
          "author_name": "alexkaggle95",
          "author_url": "",
          "post_date": "10/19/2019 09:03:00",
          "content": "<p>I don't think timezone would matter anyway as midday in for example France carries the same characteristics as midday somewhere else. I think location matters more than timezone.</p>\n<p>Besides I don't think they run experiment on sites that far from each other, and also I think algorithms would be able to handle it as each site have unique site id.</p>\n<p>Correct me if I'm wrong.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "650186": "So we have timestamp where we can extract information such as hour, month, year, weekday etc..\n\nBut does it make sense to extract things such as weekday or year?\n\nI mean we know that winter is cold so month can tell us if its winter or not.\nWe know that nights are colder than midday so hour can also have some pattern in energy consumption.\n\nBut I have seen some notebooks with weekday and year as features and to be honest.. Does it really make any sense to include those?",
    "650311": "&gt; But I have seen some notebooks with weekday and year as features and to be honest\n\nThose may help. \n\n- There are educational institutes/offices which may only operate during the weekdays (more energy consumption) and closed over the weekends (lesser energy consumption). Shopping malls may be more crowded over the weekends (Heavy energy consumption) and less crowded during the weekdays. Thus is_weekend is an important parameter. Similarly, holidays and days of yearly sales could be important features as well.\n\n- The energy consumption of a building could be increasing year over year. But, since our training data belongs to just one year, not sure how effective it will be. May be we should build a feature with number of months since 2015 Jan (or something like that to capture the long term trend in the data)\n\nFinally, one thing which I have learned about ML from Jeremy Haward (from his fast.ai course) is : \n\nQuestion : \"Will blah help as a feature?\"\nAnswer: \"Try blah &amp; blah and see...\"\n\nHope this helps.",
    "650395": "Thanks, that clears things out.",
    "650548": "Thanks arnab, for the explanations you provided. In fact, all of the above explanations are valid, since Energy is a resource that we all use, some more, some less, varying according to the days within the year. Thanks.",
    "650564": "Thanks!",
    "650863": "Yes it does. buildings metadata stores building type, so you can extract \"working_hours\" factor with lower energy demand",
    "651730": "Maybe this is a dumb question but are these timestamp in the same time zone?",
    "652703": "I don't think timezone would matter anyway as midday in for example France carries the same characteristics as midday somewhere else. I think location matters more than timezone.\n\nBesides I don't think they run experiment on sites that far from each other, and also I think algorithms would be able to handle it as each site have unique site id.\n\nCorrect me if I'm wrong."
  },
  "source": "meta"
}