{
  "id": 113270,
  "title": "Which metric is better?",
  "url": "/competitions/ashrae-energy-prediction/discussion/113270",
  "author_name": "",
  "post_date": "2019-10-18T06:52:23.412359Z",
  "votes": 4,
  "comment_count": 6,
  "views": 0,
  "content": "<p>RMSE or MAE or MSE or the other?</p>",
  "messages": [
    {
      "id": "651966",
      "postDate": "10/18/2019 06:52:23",
      "content": "<p>RMSE or MAE or MSE or the other?</p>",
      "rawMarkdown": "RMSE or MAE or MSE or the other?",
      "votes": null
    },
    {
      "id": "651974",
      "postDate": "10/18/2019 07:09:30",
      "content": "<p><a href=\"/laplaceplanet\">@laplaceplanet</a>  For this competition, I would recommend using a RMSLE loss function as it is the evaluation metric.</p>\n\n<p><strong>RMSE or MAE or MSE. Which is better?</strong>\nI think that this can be narrowed down to a comparison between RMSE and MAE.\nRMSE penalises larger errors compared to MAE. \nUltimately, the choice of metric is highly dependant on the problem you are apply the model to.</p>",
      "rawMarkdown": "laplaceplanet  For this competition, I would recommend using a RMSLE loss function as it is the evaluation metric.\n\n**RMSE or MAE or MSE. Which is better?**\nI think that this can be narrowed down to a comparison between RMSE and MAE.\nRMSE penalises larger errors compared to MAE. \nUltimately, the choice of metric is highly dependant on the problem you are apply the model to.",
      "votes": null
    },
    {
      "id": "652208",
      "postDate": "10/18/2019 14:10:46",
      "content": "<p>MAE in this case is good and still think most of the people have already correct answers.\nJust cannot make it smaller because of incomplete data.\nup</p>",
      "rawMarkdown": "MAE in this case is good and still think most of the people have already correct answers.\nJust cannot make it smaller because of incomplete data.\nup",
      "votes": null
    },
    {
      "id": "652229",
      "postDate": "10/18/2019 14:35:58",
      "content": "<p>For the purposes of setting up a competition, when your response variable is non-negative and highly skewed, you should use a Gini metric.  There is a reason why insurance competitions typically use a Gini:  it's the only practical metric that will not reward your model for giving you obviously stupid predictions in such a situation.</p>",
      "rawMarkdown": "For the purposes of setting up a competition, when your response variable is non-negative and highly skewed, you should use a Gini metric.  There is a reason why insurance competitions typically use a Gini:  it's the only practical metric that will not reward your model for giving you obviously stupid predictions in such a situation.",
      "votes": null
    },
    {
      "id": "652237",
      "postDate": "10/18/2019 14:52:12",
      "content": "<p>Thanks</p>",
      "rawMarkdown": "Thanks",
      "votes": null
    },
    {
      "id": "652238",
      "postDate": "10/18/2019 14:52:37",
      "content": "<p>you are right, thanks!</p>",
      "rawMarkdown": "you are right, thanks!",
      "votes": null
    },
    {
      "id": "652240",
      "postDate": "10/18/2019 14:53:28",
      "content": "<p>Thanks! I will find my way</p>",
      "rawMarkdown": "Thanks! I will find my way",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 651974,
      "author_name": "tiannnn",
      "author_url": "",
      "post_date": "10/18/2019 07:09:30",
      "content": "<p><a href=\"/laplaceplanet\">@laplaceplanet</a>  For this competition, I would recommend using a RMSLE loss function as it is the evaluation metric.</p>\n\n<p><strong>RMSE or MAE or MSE. Which is better?</strong>\nI think that this can be narrowed down to a comparison between RMSE and MAE.\nRMSE penalises larger errors compared to MAE. \nUltimately, the choice of metric is highly dependant on the problem you are apply the model to.</p>",
      "votes": null,
      "replies": [
        {
          "id": 652237,
          "author_name": "laplaceplanet",
          "author_url": "",
          "post_date": "10/18/2019 14:52:12",
          "content": "<p>Thanks</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 652208,
      "author_name": "zinovadr",
      "author_url": "",
      "post_date": "10/18/2019 14:10:46",
      "content": "<p>MAE in this case is good and still think most of the people have already correct answers.\nJust cannot make it smaller because of incomplete data.\nup</p>",
      "votes": null,
      "replies": [
        {
          "id": 652238,
          "author_name": "laplaceplanet",
          "author_url": "",
          "post_date": "10/18/2019 14:52:37",
          "content": "<p>you are right, thanks!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 652229,
      "author_name": "dmitriyguller",
      "author_url": "",
      "post_date": "10/18/2019 14:35:58",
      "content": "<p>For the purposes of setting up a competition, when your response variable is non-negative and highly skewed, you should use a Gini metric.  There is a reason why insurance competitions typically use a Gini:  it's the only practical metric that will not reward your model for giving you obviously stupid predictions in such a situation.</p>",
      "votes": null,
      "replies": [
        {
          "id": 652240,
          "author_name": "laplaceplanet",
          "author_url": "",
          "post_date": "10/18/2019 14:53:28",
          "content": "<p>Thanks! I will find my way</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "651966": "RMSE or MAE or MSE or the other?",
    "651974": "laplaceplanet  For this competition, I would recommend using a RMSLE loss function as it is the evaluation metric.\n\n**RMSE or MAE or MSE. Which is better?**\nI think that this can be narrowed down to a comparison between RMSE and MAE.\nRMSE penalises larger errors compared to MAE. \nUltimately, the choice of metric is highly dependant on the problem you are apply the model to.",
    "652208": "MAE in this case is good and still think most of the people have already correct answers.\nJust cannot make it smaller because of incomplete data.\nup",
    "652229": "For the purposes of setting up a competition, when your response variable is non-negative and highly skewed, you should use a Gini metric.  There is a reason why insurance competitions typically use a Gini:  it's the only practical metric that will not reward your model for giving you obviously stupid predictions in such a situation.",
    "652237": "Thanks",
    "652238": "you are right, thanks!",
    "652240": "Thanks! I will find my way"
  },
  "source": "meta"
}