{
  "id": 666129,
  "title": "Shouldn't the biomass of clover be 0 for this example test chart?",
  "url": "/competitions/csiro-biomass/discussion/666129",
  "author_name": "",
  "post_date": "2026-01-05T13:51:19.748752100Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I see that many models' inference results for this clover image are not 0. Why is that?<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F24569830%2F21817e99322d7cd85cfaa0096d6717aa%2FID1001187975.jpg?generation=1767621069052764&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "3386551",
      "postDate": "01/05/2026 13:51:19",
      "content": "<p>I see that many models' inference results for this clover image are not 0. Why is that?<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F24569830%2F21817e99322d7cd85cfaa0096d6717aa%2FID1001187975.jpg?generation=1767621069052764&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "I see that many models' inference results for this clover image are not 0. Why is that?![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F24569830%2F21817e99322d7cd85cfaa0096d6717aa%2FID1001187975.jpg?generation=1767621069052764&alt=media)",
      "votes": null
    },
    {
      "id": "3386557",
      "postDate": "01/05/2026 14:02:32",
      "content": "<p>I've observed that many models over-predict on the low end and under-predict on the high end. This is likely just a model \"hedging it's bets\" against the most extreme examples in the dataset, though I do find it a little odd given the proportion of samples with 0 clover. Different loss functions and model strategies can address this to varying extents. The split you use to train the model can also impact this.</p>",
      "rawMarkdown": "I've observed that many models over-predict on the low end and under-predict on the high end. This is likely just a model \"hedging it's bets\" against the most extreme examples in the dataset, though I do find it a little odd given the proportion of samples with 0 clover. Different loss functions and model strategies can address this to varying extents. The split you use to train the model can also impact this.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3386557,
      "author_name": "chemdatafarmer",
      "author_url": "",
      "post_date": "01/05/2026 14:02:32",
      "content": "<p>I've observed that many models over-predict on the low end and under-predict on the high end. This is likely just a model \"hedging it's bets\" against the most extreme examples in the dataset, though I do find it a little odd given the proportion of samples with 0 clover. Different loss functions and model strategies can address this to varying extents. The split you use to train the model can also impact this.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3386551": "I see that many models' inference results for this clover image are not 0. Why is that?![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F24569830%2F21817e99322d7cd85cfaa0096d6717aa%2FID1001187975.jpg?generation=1767621069052764&alt=media)",
    "3386557": "I've observed that many models over-predict on the low end and under-predict on the high end. This is likely just a model \"hedging it's bets\" against the most extreme examples in the dataset, though I do find it a little odd given the proportion of samples with 0 clover. Different loss functions and model strategies can address this to varying extents. The split you use to train the model can also impact this."
  },
  "source": "meta"
}