{
  "id": 22501,
  "title": "Calculating the Root Mean Squared Logarithmic Error",
  "url": "/competitions/grupo-bimbo-inventory-demand/discussion/22501",
  "author_name": "",
  "post_date": "2016-07-26T12:38:19.353Z",
  "votes": 1,
  "comment_count": 7,
  "views": 1107,
  "content": "<p>For those like me that didn't know, Ben Hammer has created a Metrics R package on CRAN that has a nice function to calculate the RMSLE.</p>\n\n<p><a href=\"https://cran.r-project.org/web/packages/Metrics/Metrics.pdf\">Metrics R Package</a></p>",
  "messages": [
    {
      "id": "129065",
      "postDate": "07/26/2016 12:38:19",
      "content": "<p>For those like me that didn't know, Ben Hammer has created a Metrics R package on CRAN that has a nice function to calculate the RMSLE.</p>\n\n<p><a href=\"https://cran.r-project.org/web/packages/Metrics/Metrics.pdf\">Metrics R Package</a></p>",
      "rawMarkdown": "For those like me that didn't know, Ben Hammer has created a Metrics R package on CRAN that has a nice function to calculate the RMSLE.\r\n\r\n[Metrics R Package][1]\r\n\r\n\r\n  [1]: https://cran.r-project.org/web/packages/Metrics/Metrics.pdf",
      "votes": null
    },
    {
      "id": "129186",
      "postDate": "07/27/2016 12:05:06",
      "content": "<p>It is better to create a new target as log(1 + original target) and use rmse on it IMHO....</p>",
      "rawMarkdown": "It is better to create a new target as log(1 + original target) and use rmse on it IMHO....",
      "votes": null
    },
    {
      "id": "129187",
      "postDate": "07/27/2016 12:08:55",
      "content": "<p>I read a bit about that... and saw it being used in the scripts... Do you know why it is better?</p>",
      "rawMarkdown": "I read a bit about that... and saw it being used in the scripts... Do you know why it is better?",
      "votes": null
    },
    {
      "id": "129188",
      "postDate": "07/27/2016 12:12:47",
      "content": "<p>It is mathematically equivalent if that is your question.</p>\n\n<p>But some packages, eg xgboost, provide rmse out of the box while they don't provide rmsle.  That's why I say it is better to use rmse.</p>",
      "rawMarkdown": "It is mathematically equivalent if that is your question.\r\n\r\nBut some packages, eg xgboost, provide rmse out of the box while they don't provide rmsle.  That's why I say it is better to use rmse.",
      "votes": null
    },
    {
      "id": "129190",
      "postDate": "07/27/2016 12:19:26",
      "content": "<p>Ahhh - I see!  </p>\n\n<p>Thanks CPMP.  </p>\n\n<p>Hope you are enjoying life in the top 10.  :)</p>",
      "rawMarkdown": "Ahhh - I see!  \r\n\r\nThanks CPMP.  \r\n\r\nHope you are enjoying life in the top 10.  :)",
      "votes": null
    },
    {
      "id": "129194",
      "postDate": "07/27/2016 12:48:22",
      "content": "<p>Sure, I'm happy to be in top 10, but this isn't really meaningful as the private LB uses a different data set (likely week 11 vs week 10 for public LB).  Forecasting 2 weeks ahead could be quite different from forecasting one week ahead...</p>",
      "rawMarkdown": "Sure, I'm happy to be in top 10, but this isn't really meaningful as the private LB uses a different data set (likely week 11 vs week 10 for public LB).  Forecasting 2 weeks ahead could be quite different from forecasting one week ahead...",
      "votes": null
    },
    {
      "id": "129198",
      "postDate": "07/27/2016 13:09:57",
      "content": "<p>Yup - but I'd be giddy just the same.  :)</p>",
      "rawMarkdown": "Yup - but I'd be giddy just the same.  :)",
      "votes": null
    },
    {
      "id": "129201",
      "postDate": "07/27/2016 13:39:45",
      "content": "<p>OK, I'm giddy and I'll brag about it while I can ;)</p>",
      "rawMarkdown": "OK, I'm giddy and I'll brag about it while I can ;)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 129186,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "07/27/2016 12:05:06",
      "content": "<p>It is better to create a new target as log(1 + original target) and use rmse on it IMHO....</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 129187,
      "author_name": "crimeprophet",
      "author_url": "",
      "post_date": "07/27/2016 12:08:55",
      "content": "<p>I read a bit about that... and saw it being used in the scripts... Do you know why it is better?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 129188,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "07/27/2016 12:12:47",
      "content": "<p>It is mathematically equivalent if that is your question.</p>\n\n<p>But some packages, eg xgboost, provide rmse out of the box while they don't provide rmsle.  That's why I say it is better to use rmse.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 129190,
      "author_name": "crimeprophet",
      "author_url": "",
      "post_date": "07/27/2016 12:19:26",
      "content": "<p>Ahhh - I see!  </p>\n\n<p>Thanks CPMP.  </p>\n\n<p>Hope you are enjoying life in the top 10.  :)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 129194,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "07/27/2016 12:48:22",
      "content": "<p>Sure, I'm happy to be in top 10, but this isn't really meaningful as the private LB uses a different data set (likely week 11 vs week 10 for public LB).  Forecasting 2 weeks ahead could be quite different from forecasting one week ahead...</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 129198,
      "author_name": "crimeprophet",
      "author_url": "",
      "post_date": "07/27/2016 13:09:57",
      "content": "<p>Yup - but I'd be giddy just the same.  :)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 129201,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "07/27/2016 13:39:45",
      "content": "<p>OK, I'm giddy and I'll brag about it while I can ;)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "129065": "For those like me that didn't know, Ben Hammer has created a Metrics R package on CRAN that has a nice function to calculate the RMSLE.\r\n\r\n[Metrics R Package][1]\r\n\r\n\r\n  [1]: https://cran.r-project.org/web/packages/Metrics/Metrics.pdf",
    "129186": "It is better to create a new target as log(1 + original target) and use rmse on it IMHO....",
    "129187": "I read a bit about that... and saw it being used in the scripts... Do you know why it is better?",
    "129188": "It is mathematically equivalent if that is your question.\r\n\r\nBut some packages, eg xgboost, provide rmse out of the box while they don't provide rmsle.  That's why I say it is better to use rmse.",
    "129190": "Ahhh - I see!  \r\n\r\nThanks CPMP.  \r\n\r\nHope you are enjoying life in the top 10.  :)",
    "129194": "Sure, I'm happy to be in top 10, but this isn't really meaningful as the private LB uses a different data set (likely week 11 vs week 10 for public LB).  Forecasting 2 weeks ahead could be quite different from forecasting one week ahead...",
    "129198": "Yup - but I'd be giddy just the same.  :)",
    "129201": "OK, I'm giddy and I'll brag about it while I can ;)"
  },
  "source": "meta"
}