{
  "id": 533060,
  "title": "Loss for spinal",
  "url": "/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/533060",
  "author_name": "",
  "post_date": "2024-09-09T10:53:40.509149Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>What cv loss do you have for spinal?  Do you prefer binary cross entropy (multilabel) or cross entropy (5 x 3 classes)?<br>\nFor spinal loss using patches (cross entropy with weights 1, 2 ,4) I got 0.27 +- 0.02. Moreover, I checked L1 loss for point detection and I got 0.02 (prediction is 10 pixels from ground truth on average at 512 px image)</p>",
  "messages": [
    {
      "id": "2984065",
      "postDate": "09/09/2024 10:53:40",
      "content": "<p>What cv loss do you have for spinal?  Do you prefer binary cross entropy (multilabel) or cross entropy (5 x 3 classes)?<br>\nFor spinal loss using patches (cross entropy with weights 1, 2 ,4) I got 0.27 +- 0.02. Moreover, I checked L1 loss for point detection and I got 0.02 (prediction is 10 pixels from ground truth on average at 512 px image)</p>",
      "rawMarkdown": "What cv loss do you have for spinal?  Do you prefer binary cross entropy (multilabel) or cross entropy (5 x 3 classes)?\nFor spinal loss using patches (cross entropy with weights 1, 2 ,4) I got 0.27 +- 0.02. Moreover, I checked L1 loss for point detection and I got 0.02 (prediction is 10 pixels from ground truth on average at 512 px image)",
      "votes": null
    },
    {
      "id": "2989812",
      "postDate": "09/15/2024 15:08:57",
      "content": "<p>Check this: <a href=\"https://www.kaggle.com/code/junkoda/optimize-the-evaluation-metric\" target=\"_blank\">https://www.kaggle.com/code/junkoda/optimize-the-evaluation-metric</a><br>\nThis is directly comp metric implemented for loss</p>",
      "rawMarkdown": "Check this: https://www.kaggle.com/code/junkoda/optimize-the-evaluation-metric\nThis is directly comp metric implemented for loss",
      "votes": null
    },
    {
      "id": "2989953",
      "postDate": "09/15/2024 18:02:47",
      "content": "<p>Thanks, but I will try with 2 staged solution (point detection + classification)</p>",
      "rawMarkdown": "Thanks, but I will try with 2 staged solution (point detection + classification)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2989812,
      "author_name": "kfk42kfk",
      "author_url": "",
      "post_date": "09/15/2024 15:08:57",
      "content": "<p>Check this: <a href=\"https://www.kaggle.com/code/junkoda/optimize-the-evaluation-metric\" target=\"_blank\">https://www.kaggle.com/code/junkoda/optimize-the-evaluation-metric</a><br>\nThis is directly comp metric implemented for loss</p>",
      "votes": null,
      "replies": [
        {
          "id": 2989953,
          "author_name": "jankowalski2000",
          "author_url": "",
          "post_date": "09/15/2024 18:02:47",
          "content": "<p>Thanks, but I will try with 2 staged solution (point detection + classification)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2984065": "What cv loss do you have for spinal?  Do you prefer binary cross entropy (multilabel) or cross entropy (5 x 3 classes)?\nFor spinal loss using patches (cross entropy with weights 1, 2 ,4) I got 0.27 +- 0.02. Moreover, I checked L1 loss for point detection and I got 0.02 (prediction is 10 pixels from ground truth on average at 512 px image)",
    "2989812": "Check this: https://www.kaggle.com/code/junkoda/optimize-the-evaluation-metric\nThis is directly comp metric implemented for loss",
    "2989953": "Thanks, but I will try with 2 staged solution (point detection + classification)"
  },
  "source": "meta"
}