{
  "id": 646337,
  "title": "Low RMSE is not reflective of LB score",
  "url": "/competitions/csiro-biomass/discussion/646337",
  "author_name": "",
  "post_date": "2025-11-30T16:28:02.468871300Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi Fellow Kagglers - This is my first time to participate in Kaggle contest. I have been to get my model to be trained with the RMSE score of ~11 or 12 but still the LB score from the inference is no where near in the first 200 or 100. </p>\n<p>Does it mean that the RMSE has to be even more reduced or does it mean that the inference logic is wrong ? Any idea. I know its tough to comment on this without seeing the inference or model. I have also trained the model with the external Irish and other clover datasets mentioned in the discussions but still no luck ? Can someone provide pointers on where I could be going wrong ? </p>",
  "messages": [
    {
      "id": "3355460",
      "postDate": "11/30/2025 16:28:02",
      "content": "<p>Hi Fellow Kagglers - This is my first time to participate in Kaggle contest. I have been to get my model to be trained with the RMSE score of ~11 or 12 but still the LB score from the inference is no where near in the first 200 or 100. </p>\n<p>Does it mean that the RMSE has to be even more reduced or does it mean that the inference logic is wrong ? Any idea. I know its tough to comment on this without seeing the inference or model. I have also trained the model with the external Irish and other clover datasets mentioned in the discussions but still no luck ? Can someone provide pointers on where I could be going wrong ? </p>",
      "rawMarkdown": "Hi Fellow Kagglers - This is my first time to participate in Kaggle contest. I have been to get my model to be trained with the RMSE score of ~11 or 12 but still the LB score from the inference is no where near in the first 200 or 100. \n\nDoes it mean that the RMSE has to be even more reduced or does it mean that the inference logic is wrong ? Any idea. I know its tough to comment on this without seeing the inference or model. I have also trained the model with the external Irish and other clover datasets mentioned in the discussions but still no luck ? Can someone provide pointers on where I could be going wrong ?",
      "votes": null
    },
    {
      "id": "3355979",
      "postDate": "11/30/2025 18:59:20",
      "content": "<p>The evaluation metric is weighted R2 and not RMSE</p>",
      "rawMarkdown": "The evaluation metric is weighted R2 and not RMSE",
      "votes": null
    },
    {
      "id": "3357754",
      "postDate": "12/01/2025 11:28:14",
      "content": "<p>Thanks William.. Yes I understand that ;even i measure the R2 - by passing the training set via the pipeline and calculate it against the truth - R2 scores are very good but my scores are not representative of this. </p>",
      "rawMarkdown": "Thanks William.. Yes I understand that ;even i measure the R2 - by passing the training set via the pipeline and calculate it against the truth - R2 scores are very good but my scores are not representative of this.",
      "votes": null
    },
    {
      "id": "3359765",
      "postDate": "12/01/2025 19:12:50",
      "content": "<p>You have to measure r2 against data your model hasn't seen. Or else you are getting deceptive r2 scores</p>",
      "rawMarkdown": "You have to measure r2 against data your model hasn't seen. Or else you are getting deceptive r2 scores",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3355979,
      "author_name": "williamalabi",
      "author_url": "",
      "post_date": "11/30/2025 18:59:20",
      "content": "<p>The evaluation metric is weighted R2 and not RMSE</p>",
      "votes": null,
      "replies": [
        {
          "id": 3357754,
          "author_name": "crankyoldmankk",
          "author_url": "",
          "post_date": "12/01/2025 11:28:14",
          "content": "<p>Thanks William.. Yes I understand that ;even i measure the R2 - by passing the training set via the pipeline and calculate it against the truth - R2 scores are very good but my scores are not representative of this. </p>",
          "votes": null,
          "replies": [
            {
              "id": 3359765,
              "author_name": "williamalabi",
              "author_url": "",
              "post_date": "12/01/2025 19:12:50",
              "content": "<p>You have to measure r2 against data your model hasn't seen. Or else you are getting deceptive r2 scores</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3355460": "Hi Fellow Kagglers - This is my first time to participate in Kaggle contest. I have been to get my model to be trained with the RMSE score of ~11 or 12 but still the LB score from the inference is no where near in the first 200 or 100. \n\nDoes it mean that the RMSE has to be even more reduced or does it mean that the inference logic is wrong ? Any idea. I know its tough to comment on this without seeing the inference or model. I have also trained the model with the external Irish and other clover datasets mentioned in the discussions but still no luck ? Can someone provide pointers on where I could be going wrong ?",
    "3355979": "The evaluation metric is weighted R2 and not RMSE",
    "3357754": "Thanks William.. Yes I understand that ;even i measure the R2 - by passing the training set via the pipeline and calculate it against the truth - R2 scores are very good but my scores are not representative of this.",
    "3359765": "You have to measure r2 against data your model hasn't seen. Or else you are getting deceptive r2 scores"
  },
  "source": "meta"
}