{
  "id": 66041,
  "title": "80/20 rule",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/66041",
  "author_name": "",
  "post_date": "2018-09-17T14:53:58.191273400Z",
  "votes": 4,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I have been wondering where can I observe 80/20 rule on this data. At first, I thought it only comes from the ratio of 'zero transaction revenue' to 'non-zero transaction revenue'  (98.6% : 1.4%). Later on I realize that the rule can also be seen inside 'non-zero transaction revenue' by looking at the percentile of the mean of the fullVisitorId_Sum(TransactionRevenue):</p>\n\n<p>mean             1.540688e+08 <br>\npercentile_mean    82.07283</p>\n\n<p>The calculated percentile_mean means that 82% of visitors have transaction revenue below average \nand vice versa. If you calculate total transaction for each group, you will get surprisingly close to 80:20, i. e. 82% people make 26% of total revenue, and 18% people make 74% of total revenue. If I use Log(1+TR), the data distribution changes, and 80/20 rule can't be observed any longer since the distribution changes. So it's better to extract the information out before applying any transformation.</p>",
  "messages": [
    {
      "id": "388769",
      "postDate": "09/17/2018 14:53:58",
      "content": "<p>I have been wondering where can I observe 80/20 rule on this data. At first, I thought it only comes from the ratio of 'zero transaction revenue' to 'non-zero transaction revenue'  (98.6% : 1.4%). Later on I realize that the rule can also be seen inside 'non-zero transaction revenue' by looking at the percentile of the mean of the fullVisitorId_Sum(TransactionRevenue):</p>\n\n<p>mean             1.540688e+08 <br>\npercentile_mean    82.07283</p>\n\n<p>The calculated percentile_mean means that 82% of visitors have transaction revenue below average \nand vice versa. If you calculate total transaction for each group, you will get surprisingly close to 80:20, i. e. 82% people make 26% of total revenue, and 18% people make 74% of total revenue. If I use Log(1+TR), the data distribution changes, and 80/20 rule can't be observed any longer since the distribution changes. So it's better to extract the information out before applying any transformation.</p>",
      "rawMarkdown": "I have been wondering where can I observe 80/20 rule on this data. At first, I thought it only comes from the ratio of 'zero transaction revenue' to 'non-zero transaction revenue'  (98.6% : 1.4%). Later on I realize that the rule can also be seen inside 'non-zero transaction revenue' by looking at the percentile of the mean of the fullVisitorId_Sum(TransactionRevenue):\n\nmean             1.540688e+08       \npercentile_mean    82.07283\n\nThe calculated percentile_mean means that 82% of visitors have transaction revenue below average \nand vice versa. If you calculate total transaction for each group, you will get surprisingly close to 80:20, i. e. 82% people make 26% of total revenue, and 18% people make 74% of total revenue. If I use Log(1+TR), the data distribution changes, and 80/20 rule can't be observed any longer since the distribution changes. So it's better to extract the information out before applying any transformation.",
      "votes": null
    },
    {
      "id": "388775",
      "postDate": "09/17/2018 15:10:03",
      "content": "<p>I don't think they literally meant 80/20.  The \"80/20 rule\" is a shorthand for Pareto principle, which is more general than this and doesn't require the numbers to literally be 80 and 20.  It's just that \"80/20 rule\" is more catchy than \"x/(100 - x) rule\".</p>",
      "rawMarkdown": "I don't think they literally meant 80/20.  The \"80/20 rule\" is a shorthand for Pareto principle, which is more general than this and doesn't require the numbers to literally be 80 and 20.  It's just that \"80/20 rule\" is more catchy than \"x/(100 - x) rule\".",
      "votes": null
    },
    {
      "id": "388781",
      "postDate": "09/17/2018 15:21:25",
      "content": "<p>Hi Dmitriy, thanks for the comment. Sure, I just want to mention there is a second \"80/20\" rule in the data which might be lost when someone use Log(1+TR) instead of Sum(TR) first.</p>",
      "rawMarkdown": "Hi Dmitriy, thanks for the comment. Sure, I just want to mention there is a second \"80/20\" rule in the data which might be lost when someone use Log(1+TR) instead of Sum(TR) first.",
      "votes": null
    },
    {
      "id": "388816",
      "postDate": "09/17/2018 16:27:18",
      "content": "<p>My understanding here is that maybe we need to figure out those significant 20% to produce 80% of predictive capability! :D</p>",
      "rawMarkdown": "My understanding here is that maybe we need to figure out those significant 20% to produce 80% of predictive capability! :D",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 388775,
      "author_name": "dmitriyguller",
      "author_url": "",
      "post_date": "09/17/2018 15:10:03",
      "content": "<p>I don't think they literally meant 80/20.  The \"80/20 rule\" is a shorthand for Pareto principle, which is more general than this and doesn't require the numbers to literally be 80 and 20.  It's just that \"80/20 rule\" is more catchy than \"x/(100 - x) rule\".</p>",
      "votes": null,
      "replies": [
        {
          "id": 388781,
          "author_name": "taufikadi",
          "author_url": "",
          "post_date": "09/17/2018 15:21:25",
          "content": "<p>Hi Dmitriy, thanks for the comment. Sure, I just want to mention there is a second \"80/20\" rule in the data which might be lost when someone use Log(1+TR) instead of Sum(TR) first.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 388816,
      "author_name": "",
      "author_url": "",
      "post_date": "09/17/2018 16:27:18",
      "content": "<p>My understanding here is that maybe we need to figure out those significant 20% to produce 80% of predictive capability! :D</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "388769": "I have been wondering where can I observe 80/20 rule on this data. At first, I thought it only comes from the ratio of 'zero transaction revenue' to 'non-zero transaction revenue'  (98.6% : 1.4%). Later on I realize that the rule can also be seen inside 'non-zero transaction revenue' by looking at the percentile of the mean of the fullVisitorId_Sum(TransactionRevenue):\n\nmean             1.540688e+08       \npercentile_mean    82.07283\n\nThe calculated percentile_mean means that 82% of visitors have transaction revenue below average \nand vice versa. If you calculate total transaction for each group, you will get surprisingly close to 80:20, i. e. 82% people make 26% of total revenue, and 18% people make 74% of total revenue. If I use Log(1+TR), the data distribution changes, and 80/20 rule can't be observed any longer since the distribution changes. So it's better to extract the information out before applying any transformation.",
    "388775": "I don't think they literally meant 80/20.  The \"80/20 rule\" is a shorthand for Pareto principle, which is more general than this and doesn't require the numbers to literally be 80 and 20.  It's just that \"80/20 rule\" is more catchy than \"x/(100 - x) rule\".",
    "388781": "Hi Dmitriy, thanks for the comment. Sure, I just want to mention there is a second \"80/20\" rule in the data which might be lost when someone use Log(1+TR) instead of Sum(TR) first.",
    "388816": "My understanding here is that maybe we need to figure out those significant 20% to produce 80% of predictive capability! :D"
  },
  "source": "meta"
}