{
  "id": 53817,
  "title": "Just curious how this model is to be applied within business system",
  "url": "/competitions/talkingdata-adtracking-fraud-detection/discussion/53817",
  "author_name": "Validation",
  "post_date": "2018-04-05T14:02:56.603000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>The description puts </p>\n\n<blockquote>\n  <p>you’re challenged to build an algorithm that predicts whether a user\n  will download an app after clicking a mobile app ad</p>\n</blockquote>\n\n<p>Then where the model is to go, how does it work in TalkingData's business system? Deciding whether an ad will be pushed to a specific user event? </p>\n\n<p>PS: The idea arising from many pretty influential \"future features\" , namely count/frequency features, which are always used in ML. Eg, count(ip_day_hour), count(ip_device)</p>\n\n<p>Does anyone have not used any of these features? Or someone else treated this competition as a time-series one, using historical rolling mean/std features instead?</p>",
  "messages": [
    {
      "id": 309518,
      "postDate": "2018-04-05T14:02:56.603Z",
      "content": "<p>The description puts </p>\n\n<blockquote>\n  <p>you’re challenged to build an algorithm that predicts whether a user\n  will download an app after clicking a mobile app ad</p>\n</blockquote>\n\n<p>Then where the model is to go, how does it work in TalkingData's business system? Deciding whether an ad will be pushed to a specific user event? </p>\n\n<p>PS: The idea arising from many pretty influential \"future features\" , namely count/frequency features, which are always used in ML. Eg, count(ip_day_hour), count(ip_device)</p>\n\n<p>Does anyone have not used any of these features? Or someone else treated this competition as a time-series one, using historical rolling mean/std features instead?</p>",
      "rawMarkdown": "The description puts \n\n&gt; you’re challenged to build an algorithm that predicts whether a user\n&gt; will download an app after clicking a mobile app ad\n\nThen where the model is to go, how does it work in TalkingData's business system? Deciding whether an ad will be pushed to a specific user event? \n\nPS: The idea arising from many pretty influential \"future features\" , namely count/frequency features, which are always used in ML. Eg, count(ip_day_hour), count(ip_device)\n\nDoes anyone have not used any of these features? Or someone else treated this competition as a time-series one, using historical rolling mean/std features instead?",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "309518": "The description puts \n\n&gt; you’re challenged to build an algorithm that predicts whether a user\n&gt; will download an app after clicking a mobile app ad\n\nThen where the model is to go, how does it work in TalkingData's business system? Deciding whether an ad will be pushed to a specific user event? \n\nPS: The idea arising from many pretty influential \"future features\" , namely count/frequency features, which are always used in ML. Eg, count(ip_day_hour), count(ip_device)\n\nDoes anyone have not used any of these features? Or someone else treated this competition as a time-series one, using historical rolling mean/std features instead?"
  }
}