{
  "id": 633377,
  "title": "Weighted R-squared metric clarification",
  "url": "/competitions/csiro-biomass/discussion/633377",
  "author_name": "",
  "post_date": "2025-11-19T01:24:32.571724700Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Just wanted to ask about the disparity between the metric used in the paper and the metric shown in the overview section of the competition.</p>\n<p>In the paper the metric is the weighted average of the R-squared values computed independently for each target type whereas in the overview the R-squared metric is computed for all target values together with weighting applied both in the calculation of the mean target as well as both sum of squares.</p>\n<p>There have been some discussions that allude to the version in the paper being appropriate but that conflicts with the overview description. Also, the paper states there is a log transform applied but that is not described as part of the competition evaluation process (or at least I didn't see it).</p>\n<p>Hoping to get some confirmation on what evaluation process is actually used as otherwise I assume it is exactly as described here on Kaggle in the overview section. Thanks!</p>",
  "messages": [
    {
      "id": "3339234",
      "postDate": "11/19/2025 01:24:32",
      "content": "<p>Just wanted to ask about the disparity between the metric used in the paper and the metric shown in the overview section of the competition.</p>\n<p>In the paper the metric is the weighted average of the R-squared values computed independently for each target type whereas in the overview the R-squared metric is computed for all target values together with weighting applied both in the calculation of the mean target as well as both sum of squares.</p>\n<p>There have been some discussions that allude to the version in the paper being appropriate but that conflicts with the overview description. Also, the paper states there is a log transform applied but that is not described as part of the competition evaluation process (or at least I didn't see it).</p>\n<p>Hoping to get some confirmation on what evaluation process is actually used as otherwise I assume it is exactly as described here on Kaggle in the overview section. Thanks!</p>",
      "rawMarkdown": "Just wanted to ask about the disparity between the metric used in the paper and the metric shown in the overview section of the competition.\n\nIn the paper the metric is the weighted average of the R-squared values computed independently for each target type whereas in the overview the R-squared metric is computed for all target values together with weighting applied both in the calculation of the mean target as well as both sum of squares.\n\nThere have been some discussions that allude to the version in the paper being appropriate but that conflicts with the overview description. Also, the paper states there is a log transform applied but that is not described as part of the competition evaluation process (or at least I didn't see it).\n\nHoping to get some confirmation on what evaluation process is actually used as otherwise I assume it is exactly as described here on Kaggle in the overview section. Thanks!",
      "votes": null
    },
    {
      "id": "3339246",
      "postDate": "11/19/2025 01:43:04",
      "content": "<p>Subject to the description on the overview page</p>",
      "rawMarkdown": "Subject to the description on the overview page",
      "votes": null
    },
    {
      "id": "3339293",
      "postDate": "11/19/2025 02:33:41",
      "content": "<p>Thanks Qiyu!</p>",
      "rawMarkdown": "Thanks Qiyu!",
      "votes": null
    },
    {
      "id": "3339555",
      "postDate": "11/19/2025 05:50:43",
      "content": "<p>No worries!😀</p>",
      "rawMarkdown": "No worries!😀",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3339246,
      "author_name": "qiyuliaocsiro",
      "author_url": "",
      "post_date": "11/19/2025 01:43:04",
      "content": "<p>Subject to the description on the overview page</p>",
      "votes": null,
      "replies": [
        {
          "id": 3339293,
          "author_name": "bklynch",
          "author_url": "",
          "post_date": "11/19/2025 02:33:41",
          "content": "<p>Thanks Qiyu!</p>",
          "votes": null,
          "replies": [
            {
              "id": 3339555,
              "author_name": "qiyuliaocsiro",
              "author_url": "",
              "post_date": "11/19/2025 05:50:43",
              "content": "<p>No worries!😀</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3339234": "Just wanted to ask about the disparity between the metric used in the paper and the metric shown in the overview section of the competition.\n\nIn the paper the metric is the weighted average of the R-squared values computed independently for each target type whereas in the overview the R-squared metric is computed for all target values together with weighting applied both in the calculation of the mean target as well as both sum of squares.\n\nThere have been some discussions that allude to the version in the paper being appropriate but that conflicts with the overview description. Also, the paper states there is a log transform applied but that is not described as part of the competition evaluation process (or at least I didn't see it).\n\nHoping to get some confirmation on what evaluation process is actually used as otherwise I assume it is exactly as described here on Kaggle in the overview section. Thanks!",
    "3339246": "Subject to the description on the overview page",
    "3339293": "Thanks Qiyu!",
    "3339555": "No worries!😀"
  },
  "source": "meta"
}