{
  "id": 66202,
  "title": "Advantage of using natural log?",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/66202",
  "author_name": "Ranjeet Jain",
  "post_date": "2018-09-19T06:50:20.924000",
  "votes": 0,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Many of us will be having the same question. </p>\n\n<p>What is the advantage of using natural log(revenue) over revenue.</p>\n\n<p>I am new to this just want to know the differences and the impact it will make? </p>\n\n<p>and when to use log (something) over something </p>\n\n<p>PS - new to data science field. Just want to understand the basics.</p>",
  "messages": [
    {
      "id": 390329,
      "postDate": "2018-09-20T03:21:05.470Z",
      "content": "<p>It changes the distribution.  Revenue often (and in this case too, I think) sometimes follows something that (at least for a reasonable range) looks similar to a  \"log normal\" distribution, which looks normal-ish but with a long tail to the right.  Taking logs will make it closer to \"normal\" so, for example, linear regression (and anything that uses linear regression as a substep) works better.    I learned about log normal distributions myself from \"Data Analysis with Open Source Tools: A Hands-On Guide for Programmers and Data Scientists\" by Philipp K. Janert, an O'Reilly book.</p>\n\n<p>It definitely helped in the house prices advanced regression techniques  \"getting started\" competition, though you had to take the logs yourself, then exponentiate the result.</p>",
      "rawMarkdown": "It changes the distribution.  Revenue often (and in this case too, I think) sometimes follows something that (at least for a reasonable range) looks similar to a  \"log normal\" distribution, which looks normal-ish but with a long tail to the right.  Taking logs will make it closer to \"normal\" so, for example, linear regression (and anything that uses linear regression as a substep) works better.    I learned about log normal distributions myself from \"Data Analysis with Open Source Tools: A Hands-On Guide for Programmers and Data Scientists\" by Philipp K. Janert, an O'Reilly book.\n\nIt definitely helped in the house prices advanced regression techniques  \"getting started\" competition, though you had to take the logs yourself, then exponentiate the result.",
      "votes": 2
    },
    {
      "id": 389753,
      "postDate": "2018-09-19T06:50:20.923Z",
      "content": "<p>Many of us will be having the same question. </p>\n\n<p>What is the advantage of using natural log(revenue) over revenue.</p>\n\n<p>I am new to this just want to know the differences and the impact it will make? </p>\n\n<p>and when to use log (something) over something </p>\n\n<p>PS - new to data science field. Just want to understand the basics.</p>",
      "rawMarkdown": "Many of us will be having the same question. \n\nWhat is the advantage of using natural log(revenue) over revenue.\n\nI am new to this just want to know the differences and the impact it will make? \n\nand when to use log (something) over something \n\nPS - new to data science field. Just want to understand the basics."
    },
    {
      "id": 389783,
      "postDate": "2018-09-19T08:12:43.943Z",
      "rawMarkdown": "",
      "votes": 4,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 390329,
      "author_name": "Todd D. Vance",
      "author_url": "",
      "post_date": "2018-09-20T03:21:05.470000",
      "content": "<p>It changes the distribution.  Revenue often (and in this case too, I think) sometimes follows something that (at least for a reasonable range) looks similar to a  \"log normal\" distribution, which looks normal-ish but with a long tail to the right.  Taking logs will make it closer to \"normal\" so, for example, linear regression (and anything that uses linear regression as a substep) works better.    I learned about log normal distributions myself from \"Data Analysis with Open Source Tools: A Hands-On Guide for Programmers and Data Scientists\" by Philipp K. Janert, an O'Reilly book.</p>\n\n<p>It definitely helped in the house prices advanced regression techniques  \"getting started\" competition, though you had to take the logs yourself, then exponentiate the result.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 389783,
      "author_name": "",
      "author_url": "",
      "post_date": "2018-09-19T08:12:43.943000",
      "content": "",
      "votes": 4,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "390329": "It changes the distribution.  Revenue often (and in this case too, I think) sometimes follows something that (at least for a reasonable range) looks similar to a  \"log normal\" distribution, which looks normal-ish but with a long tail to the right.  Taking logs will make it closer to \"normal\" so, for example, linear regression (and anything that uses linear regression as a substep) works better.    I learned about log normal distributions myself from \"Data Analysis with Open Source Tools: A Hands-On Guide for Programmers and Data Scientists\" by Philipp K. Janert, an O'Reilly book.\n\nIt definitely helped in the house prices advanced regression techniques  \"getting started\" competition, though you had to take the logs yourself, then exponentiate the result.",
    "389753": "Many of us will be having the same question. \n\nWhat is the advantage of using natural log(revenue) over revenue.\n\nI am new to this just want to know the differences and the impact it will make? \n\nand when to use log (something) over something \n\nPS - new to data science field. Just want to understand the basics.",
    "389783": ""
  }
}