{
  "id": 667570,
  "title": "Are certain targets overfitting while the rest are not?",
  "url": "/competitions/csiro-biomass/discussion/667570",
  "author_name": "",
  "post_date": "2026-01-13T06:52:45.884332Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Epoch X | Train: 8.4632 | Val: 8.0946 | LR: 5.00e-04</p>\n<ul>\n<li>Train R²: 0.6572 | Green:0.766 | Dead:0.367 | Clover:0.728 | GDM:0.724 | Total:0.652</li>\n<li>Val   R²: 0.6857 | Green:0.666 | Dead:0.513 | Clover:0.539 | GDM:0.760 | Total:0.724</li>\n</ul>\n<p>Should I interpret the above result as \"<strong>The targets Green and Clover are overfitting while the rest are not!</strong>\" ?</p>\n<p><strong>Note :</strong> that this same pattern is seen in all the surrounding epochs consistently and is not a result of just some oscillation.</p>\n<p>If the above is interpretation is correct, should we add more restrictions to Green and Clover (maybe by adding more regularization or reducing the complexity for them) in order to try getting better performance?\nOne thing to make sure during this experimentation is surely to put a constant value in the rest of the targets while making a submission.</p>",
  "messages": [
    {
      "id": "3390446",
      "postDate": "01/13/2026 06:52:45",
      "content": "<p>Epoch X | Train: 8.4632 | Val: 8.0946 | LR: 5.00e-04</p>\n<ul>\n<li>Train R²: 0.6572 | Green:0.766 | Dead:0.367 | Clover:0.728 | GDM:0.724 | Total:0.652</li>\n<li>Val   R²: 0.6857 | Green:0.666 | Dead:0.513 | Clover:0.539 | GDM:0.760 | Total:0.724</li>\n</ul>\n<p>Should I interpret the above result as \"<strong>The targets Green and Clover are overfitting while the rest are not!</strong>\" ?</p>\n<p><strong>Note :</strong> that this same pattern is seen in all the surrounding epochs consistently and is not a result of just some oscillation.</p>\n<p>If the above is interpretation is correct, should we add more restrictions to Green and Clover (maybe by adding more regularization or reducing the complexity for them) in order to try getting better performance?\nOne thing to make sure during this experimentation is surely to put a constant value in the rest of the targets while making a submission.</p>",
      "rawMarkdown": "Epoch X | Train: 8.4632 | Val: 8.0946 | LR: 5.00e-04\n-  Train R²: 0.6572 | Green:0.766 | Dead:0.367 | Clover:0.728 | GDM:0.724 | Total:0.652\n-  Val   R²: 0.6857 | Green:0.666 | Dead:0.513 | Clover:0.539 | GDM:0.760 | Total:0.724\n\nShould I interpret the above result as \"**The targets Green and Clover are overfitting while the rest are not!**\" ?\n\n**Note :** that this same pattern is seen in all the surrounding epochs consistently and is not a result of just some oscillation.\n\nIf the above is interpretation is correct, should we add more restrictions to Green and Clover (maybe by adding more regularization or reducing the complexity for them) in order to try getting better performance?\nOne thing to make sure during this experimentation is surely to put a constant value in the rest of the targets while making a submission.",
      "votes": null
    },
    {
      "id": "3390698",
      "postDate": "01/13/2026 18:51:18",
      "content": "<p>No, it is problem of a small dataset and poor train val split.</p>",
      "rawMarkdown": "No, it is problem of a small dataset and poor train val split.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3390698,
      "author_name": "jankowalski2000",
      "author_url": "",
      "post_date": "01/13/2026 18:51:18",
      "content": "<p>No, it is problem of a small dataset and poor train val split.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3390446": "Epoch X | Train: 8.4632 | Val: 8.0946 | LR: 5.00e-04\n-  Train R²: 0.6572 | Green:0.766 | Dead:0.367 | Clover:0.728 | GDM:0.724 | Total:0.652\n-  Val   R²: 0.6857 | Green:0.666 | Dead:0.513 | Clover:0.539 | GDM:0.760 | Total:0.724\n\nShould I interpret the above result as \"**The targets Green and Clover are overfitting while the rest are not!**\" ?\n\n**Note :** that this same pattern is seen in all the surrounding epochs consistently and is not a result of just some oscillation.\n\nIf the above is interpretation is correct, should we add more restrictions to Green and Clover (maybe by adding more regularization or reducing the complexity for them) in order to try getting better performance?\nOne thing to make sure during this experimentation is surely to put a constant value in the rest of the targets while making a submission.",
    "3390698": "No, it is problem of a small dataset and poor train val split."
  },
  "source": "meta"
}