{
  "id": 530282,
  "title": "Accuracies Across Labels",
  "url": "/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/530282",
  "author_name": "",
  "post_date": "2024-08-25T17:59:52.779778100Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>What are the accuracies that you get on train data?</p>\n<p>Have people monitored them across these 4-categories.</p>\n<ol>\n<li>spinal_canal_stenosis</li>\n<li>neural_foraminal_narrowing</li>\n<li>subarticular_stenosis</li>\n<li>spinal_any_severe</li>\n</ol>\n<p>or have you generally monitored overall accuracy?</p>\n<p>I cannot comprehend CV results and would like to monitor something I am more familiar with like accuracy.</p>\n<p>I would like to know what accuracy is good accuracy for this competition ?</p>\n<p>So that I have an ballpark idea that solution is somewhat good once it achieves this much accuracy and I can target that.</p>",
  "messages": [
    {
      "id": "2970108",
      "postDate": "08/25/2024 17:59:52",
      "content": "<p>What are the accuracies that you get on train data?</p>\n<p>Have people monitored them across these 4-categories.</p>\n<ol>\n<li>spinal_canal_stenosis</li>\n<li>neural_foraminal_narrowing</li>\n<li>subarticular_stenosis</li>\n<li>spinal_any_severe</li>\n</ol>\n<p>or have you generally monitored overall accuracy?</p>\n<p>I cannot comprehend CV results and would like to monitor something I am more familiar with like accuracy.</p>\n<p>I would like to know what accuracy is good accuracy for this competition ?</p>\n<p>So that I have an ballpark idea that solution is somewhat good once it achieves this much accuracy and I can target that.</p>",
      "rawMarkdown": "What are the accuracies that you get on train data?\n\nHave people monitored them across these 4-categories.\n\n1. spinal_canal_stenosis\n2. neural_foraminal_narrowing\n3. subarticular_stenosis\n4. spinal_any_severe\n\nor have you generally monitored overall accuracy?\n\nI cannot comprehend CV results and would like to monitor something I am more familiar with like accuracy.\n\nI would like to know what accuracy is good accuracy for this competition ?\n\nSo that I have an ballpark idea that solution is somewhat good once it achieves this much accuracy and I can target that.",
      "votes": null
    },
    {
      "id": "2970201",
      "postDate": "08/25/2024 20:10:47",
      "content": "<p>You gotta make a peace with log loss, even if it feels complicated for you. We got imbalanced dataset, where the most of the labels are <code>Normal/Mild</code> condition (77.36%). If you mark everything as 'Normal/Mild' you will get ~80% accuracy. it is not very smart thing to monitor.</p>\n<p>You can log precision, recall, f1 and so on for each of the labels if you are really up to understanding how good your solution is.</p>",
      "rawMarkdown": "You gotta make a peace with log loss, even if it feels complicated for you. We got imbalanced dataset, where the most of the labels are `Normal/Mild` condition (77.36%). If you mark everything as 'Normal/Mild' you will get ~80% accuracy. it is not very smart thing to monitor.\n\nYou can log precision, recall, f1 and so on for each of the labels if you are really up to understanding how good your solution is.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2970201,
      "author_name": "sergiosaharovskiy",
      "author_url": "",
      "post_date": "08/25/2024 20:10:47",
      "content": "<p>You gotta make a peace with log loss, even if it feels complicated for you. We got imbalanced dataset, where the most of the labels are <code>Normal/Mild</code> condition (77.36%). If you mark everything as 'Normal/Mild' you will get ~80% accuracy. it is not very smart thing to monitor.</p>\n<p>You can log precision, recall, f1 and so on for each of the labels if you are really up to understanding how good your solution is.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2970108": "What are the accuracies that you get on train data?\n\nHave people monitored them across these 4-categories.\n\n1. spinal_canal_stenosis\n2. neural_foraminal_narrowing\n3. subarticular_stenosis\n4. spinal_any_severe\n\nor have you generally monitored overall accuracy?\n\nI cannot comprehend CV results and would like to monitor something I am more familiar with like accuracy.\n\nI would like to know what accuracy is good accuracy for this competition ?\n\nSo that I have an ballpark idea that solution is somewhat good once it achieves this much accuracy and I can target that.",
    "2970201": "You gotta make a peace with log loss, even if it feels complicated for you. We got imbalanced dataset, where the most of the labels are `Normal/Mild` condition (77.36%). If you mark everything as 'Normal/Mild' you will get ~80% accuracy. it is not very smart thing to monitor.\n\nYou can log precision, recall, f1 and so on for each of the labels if you are really up to understanding how good your solution is."
  },
  "source": "meta"
}