{
  "id": 65971,
  "title": "General strategy - aggregate n' predict or predict n' aggregate?",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/65971",
  "author_name": "",
  "post_date": "2018-09-16T19:25:33.562095100Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hey everyone - I just wanted your thoughts on the general strategy for building the model in this competition. Most kernels I see first make a prediction at the session level, and then aggregate those results at the <code>fullVisitorId</code> level. Is anyone seeing good results the other way round - feature creation at the <code>fullVisitorId</code> level and then building the model at that level?</p>",
  "messages": [
    {
      "id": "388325",
      "postDate": "09/16/2018 19:25:33",
      "content": "<p>Hey everyone - I just wanted your thoughts on the general strategy for building the model in this competition. Most kernels I see first make a prediction at the session level, and then aggregate those results at the <code>fullVisitorId</code> level. Is anyone seeing good results the other way round - feature creation at the <code>fullVisitorId</code> level and then building the model at that level?</p>",
      "rawMarkdown": "Hey everyone - I just wanted your thoughts on the general strategy for building the model in this competition. Most kernels I see first make a prediction at the session level, and then aggregate those results at the `fullVisitorId` level. Is anyone seeing good results the other way round - feature creation at the `fullVisitorId` level and then building the model at that level?",
      "votes": null
    },
    {
      "id": "388330",
      "postDate": "09/16/2018 19:43:46",
      "content": "<p>Here you go :) <a href=\"https://www.kaggle.com/c/google-analytics-customer-revenue-prediction/discussion/65952\">https://www.kaggle.com/c/google-analytics-customer-revenue-prediction/discussion/65952</a></p>",
      "rawMarkdown": "Here you go :) https://www.kaggle.com/c/google-analytics-customer-revenue-prediction/discussion/65952",
      "votes": null
    },
    {
      "id": "388858",
      "postDate": "09/17/2018 17:53:21",
      "content": "<p>Thank you. That was helpful! :)</p>",
      "rawMarkdown": "Thank you. That was helpful! :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 388330,
      "author_name": "artgor",
      "author_url": "",
      "post_date": "09/16/2018 19:43:46",
      "content": "<p>Here you go :) <a href=\"https://www.kaggle.com/c/google-analytics-customer-revenue-prediction/discussion/65952\">https://www.kaggle.com/c/google-analytics-customer-revenue-prediction/discussion/65952</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 388858,
          "author_name": "surag0107",
          "author_url": "",
          "post_date": "09/17/2018 17:53:21",
          "content": "<p>Thank you. That was helpful! :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "388325": "Hey everyone - I just wanted your thoughts on the general strategy for building the model in this competition. Most kernels I see first make a prediction at the session level, and then aggregate those results at the `fullVisitorId` level. Is anyone seeing good results the other way round - feature creation at the `fullVisitorId` level and then building the model at that level?",
    "388330": "Here you go :) https://www.kaggle.com/c/google-analytics-customer-revenue-prediction/discussion/65952",
    "388858": "Thank you. That was helpful! :)"
  },
  "source": "meta"
}