{
  "id": 71187,
  "title": "This revised one is weird...",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/71187",
  "author_name": "InWook",
  "post_date": "2018-11-11T05:59:28.970000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Sorry for my poor English cuz I am a Korean...</p>\n\n<p>As I understand, this revised one is very different from the previous version. In the previous one, we know all the variable information including pageviews, referralpath, source, device... to predict the testset.  So I could sum all revenues per sessionId and get total revenue per fullvisitorId.</p>\n\n<p>But in this one, we don't know about any information for each sessionID, and we need to predict future total revenue per person without any input data at that time. For doing this, we need more information about each person, I mean we need more rows per each fullvisitorId, because we don't have any input data for future data(2018.12~2019.01) such as pageviews, devices etc. </p>\n\n<p>In this situation, I think we can only use very a few fixed information such as a country variable and we should estimate the averaged behavior of each person from the rows in the same fullvisitorId. I think this problem is very weird and nonsense because there are not enough rows in the same fullvisitorId to predict averaged behavior of each person and it is also very different from the previous competiton. So many things that I have done for the previous data would not work for this revised one, and I should begin new analysis within just 20days. </p>\n\n<p>Do I think wrong? I just wonder what other people think about this. </p>",
  "messages": [
    {
      "id": 419040,
      "postDate": "2018-11-11T05:59:28.970Z",
      "content": "<p>Sorry for my poor English cuz I am a Korean...</p>\n\n<p>As I understand, this revised one is very different from the previous version. In the previous one, we know all the variable information including pageviews, referralpath, source, device... to predict the testset.  So I could sum all revenues per sessionId and get total revenue per fullvisitorId.</p>\n\n<p>But in this one, we don't know about any information for each sessionID, and we need to predict future total revenue per person without any input data at that time. For doing this, we need more information about each person, I mean we need more rows per each fullvisitorId, because we don't have any input data for future data(2018.12~2019.01) such as pageviews, devices etc. </p>\n\n<p>In this situation, I think we can only use very a few fixed information such as a country variable and we should estimate the averaged behavior of each person from the rows in the same fullvisitorId. I think this problem is very weird and nonsense because there are not enough rows in the same fullvisitorId to predict averaged behavior of each person and it is also very different from the previous competiton. So many things that I have done for the previous data would not work for this revised one, and I should begin new analysis within just 20days. </p>\n\n<p>Do I think wrong? I just wonder what other people think about this. </p>",
      "rawMarkdown": "Sorry for my poor English cuz I am a Korean...\n\nAs I understand, this revised one is very different from the previous version. In the previous one, we know all the variable information including pageviews, referralpath, source, device... to predict the testset.  So I could sum all revenues per sessionId and get total revenue per fullvisitorId.\n\nBut in this one, we don't know about any information for each sessionID, and we need to predict future total revenue per person without any input data at that time. For doing this, we need more information about each person, I mean we need more rows per each fullvisitorId, because we don't have any input data for future data(2018.12~2019.01) such as pageviews, devices etc. \n\nIn this situation, I think we can only use very a few fixed information such as a country variable and we should estimate the averaged behavior of each person from the rows in the same fullvisitorId. I think this problem is very weird and nonsense because there are not enough rows in the same fullvisitorId to predict averaged behavior of each person and it is also very different from the previous competiton. So many things that I have done for the previous data would not work for this revised one, and I should begin new analysis within just 20days. \n\nDo I think wrong? I just wonder what other people think about this. ",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "419040": "Sorry for my poor English cuz I am a Korean...\n\nAs I understand, this revised one is very different from the previous version. In the previous one, we know all the variable information including pageviews, referralpath, source, device... to predict the testset.  So I could sum all revenues per sessionId and get total revenue per fullvisitorId.\n\nBut in this one, we don't know about any information for each sessionID, and we need to predict future total revenue per person without any input data at that time. For doing this, we need more information about each person, I mean we need more rows per each fullvisitorId, because we don't have any input data for future data(2018.12~2019.01) such as pageviews, devices etc. \n\nIn this situation, I think we can only use very a few fixed information such as a country variable and we should estimate the averaged behavior of each person from the rows in the same fullvisitorId. I think this problem is very weird and nonsense because there are not enough rows in the same fullvisitorId to predict averaged behavior of each person and it is also very different from the previous competiton. So many things that I have done for the previous data would not work for this revised one, and I should begin new analysis within just 20days. \n\nDo I think wrong? I just wonder what other people think about this. "
  }
}