{
  "id": 20938,
  "title": "Is it possible to add extra information?",
  "url": "/competitions/facebook-v-predicting-check-ins/discussion/20938",
  "author_name": "",
  "post_date": "2016-05-13T16:24:08.937Z",
  "votes": 1,
  "comment_count": 1,
  "views": 424,
  "content": "<p>I'm new on this... so I have to ask...</p>\n\n<p>Is it possible to add extra information to your model? For example more data about the events...</p>\n\n<p>Thank you for answering...</p>",
  "messages": [
    {
      "id": "119899",
      "postDate": "05/13/2016 16:24:08",
      "content": "<p>I'm new on this... so I have to ask...</p>\n\n<p>Is it possible to add extra information to your model? For example more data about the events...</p>\n\n<p>Thank you for answering...</p>",
      "rawMarkdown": "I'm new on this... so I have to ask...\r\n\r\nIs it possible to add extra information to your model? For example more data about the events...\r\n\r\nThank you for answering...",
      "votes": null
    },
    {
      "id": "119900",
      "postDate": "05/13/2016 16:31:58",
      "content": "<p>I'm new too. I'm not sure about rules for outside information, but I think you're probably allowed to incorporate whatever you'd like.</p>\n\n<p>I've been reading about &quot;feature engineering&quot;. For example, instead of feeding &quot;x&quot; into a model, maybe you also feed &quot;x^2&quot;, or perhaps turn &quot;x&quot; into a binary classification, where x=z gets a 1. </p>\n\n<p>The overall idea is that you can transform the raw 'predictors' into something that's more meaningful/ targeted at explaining the response variable.</p>\n\n<p>In the current problem, if we can figure out what time stamps are associated with a weekend, maybe we'd want to create a binary variable that's a 1 if the time corresponds to a weekend, and 0 if a weekday. I think this is why you see so many people speculating about the true units of the time vector (if you know the units, you can infer these features w/o having to actually find them in the data).</p>\n\n<p>But again, I'm new, and this is just something I've been reading about.</p>",
      "rawMarkdown": "I'm new too. I'm not sure about rules for outside information, but I think you're probably allowed to incorporate whatever you'd like.\r\n\r\nI've been reading about \"feature engineering\". For example, instead of feeding \"x\" into a model, maybe you also feed \"x^2\", or perhaps turn \"x\" into a binary classification, where x<z gets a 0, and x>=z gets a 1. \r\n\r\nThe overall idea is that you can transform the raw 'predictors' into something that's more meaningful/ targeted at explaining the response variable.\r\n\r\nIn the current problem, if we can figure out what time stamps are associated with a weekend, maybe we'd want to create a binary variable that's a 1 if the time corresponds to a weekend, and 0 if a weekday. I think this is why you see so many people speculating about the true units of the time vector (if you know the units, you can infer these features w/o having to actually find them in the data).\r\n\r\nBut again, I'm new, and this is just something I've been reading about.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 119900,
      "author_name": "zeraes",
      "author_url": "",
      "post_date": "05/13/2016 16:31:58",
      "content": "<p>I'm new too. I'm not sure about rules for outside information, but I think you're probably allowed to incorporate whatever you'd like.</p>\n\n<p>I've been reading about &quot;feature engineering&quot;. For example, instead of feeding &quot;x&quot; into a model, maybe you also feed &quot;x^2&quot;, or perhaps turn &quot;x&quot; into a binary classification, where x=z gets a 1. </p>\n\n<p>The overall idea is that you can transform the raw 'predictors' into something that's more meaningful/ targeted at explaining the response variable.</p>\n\n<p>In the current problem, if we can figure out what time stamps are associated with a weekend, maybe we'd want to create a binary variable that's a 1 if the time corresponds to a weekend, and 0 if a weekday. I think this is why you see so many people speculating about the true units of the time vector (if you know the units, you can infer these features w/o having to actually find them in the data).</p>\n\n<p>But again, I'm new, and this is just something I've been reading about.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "119899": "I'm new on this... so I have to ask...\r\n\r\nIs it possible to add extra information to your model? For example more data about the events...\r\n\r\nThank you for answering...",
    "119900": "I'm new too. I'm not sure about rules for outside information, but I think you're probably allowed to incorporate whatever you'd like.\r\n\r\nI've been reading about \"feature engineering\". For example, instead of feeding \"x\" into a model, maybe you also feed \"x^2\", or perhaps turn \"x\" into a binary classification, where x<z gets a 0, and x>=z gets a 1. \r\n\r\nThe overall idea is that you can transform the raw 'predictors' into something that's more meaningful/ targeted at explaining the response variable.\r\n\r\nIn the current problem, if we can figure out what time stamps are associated with a weekend, maybe we'd want to create a binary variable that's a 1 if the time corresponds to a weekend, and 0 if a weekday. I think this is why you see so many people speculating about the true units of the time vector (if you know the units, you can infer these features w/o having to actually find them in the data).\r\n\r\nBut again, I'm new, and this is just something I've been reading about."
  },
  "source": "meta"
}