{
  "id": 533413,
  "title": "Understanding competition",
  "url": "/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/533413",
  "author_name": "",
  "post_date": "2024-09-11T04:27:19.167853100Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hey, I'm trying to wrap my head around this competition, but I'm having a hard time doing so since I'm new to this. Here's what I've understood so far:</p>\n<ol>\n<li>We need to build a model that classifies Foraminal Narrowing, Subarticular Stenosis, and Canal Stenosis at each spinal level.</li>\n<li>Once we’ve built the model, we either submit the model itself or just the output predictions.</li>\n<li>After submission, how will the evaluation be done? Will they run my model on their dataset and calculate the score from the results? If that's the case, do I need to adjust my model to ensure it identifies the best images where the spine is most visible?</li>\n</ol>",
  "messages": [
    {
      "id": "2985858",
      "postDate": "09/11/2024 04:27:19",
      "content": "<p>Hey, I'm trying to wrap my head around this competition, but I'm having a hard time doing so since I'm new to this. Here's what I've understood so far:</p>\n<ol>\n<li>We need to build a model that classifies Foraminal Narrowing, Subarticular Stenosis, and Canal Stenosis at each spinal level.</li>\n<li>Once we’ve built the model, we either submit the model itself or just the output predictions.</li>\n<li>After submission, how will the evaluation be done? Will they run my model on their dataset and calculate the score from the results? If that's the case, do I need to adjust my model to ensure it identifies the best images where the spine is most visible?</li>\n</ol>",
      "rawMarkdown": "Hey, I'm trying to wrap my head around this competition, but I'm having a hard time doing so since I'm new to this. Here's what I've understood so far:\n\n1. We need to build a model that classifies Foraminal Narrowing, Subarticular Stenosis, and Canal Stenosis at each spinal level.\n2. Once we’ve built the model, we either submit the model itself or just the output predictions.\n3. After submission, how will the evaluation be done? Will they run my model on their dataset and calculate the score from the results? If that's the case, do I need to adjust my model to ensure it identifies the best images where the spine is most visible?",
      "votes": null
    },
    {
      "id": "2985960",
      "postDate": "09/11/2024 07:33:36",
      "content": "<ol>\n<li>Yes. But Foraminal narrowing and subarticular stenosis you have to predict for both sides - for left and right.</li>\n</ol>\n<p>So you have 5 spine levels \"l1_l2\", \"l2_l3\", \"l3_l4\", \"l4-l5, \"l5-s1\" for each level you need to predict 5 conditions: \"left foraminal narrowing\", \"left subarticular stenosis\", \"canal stenosis\", \"right subarticular stenosis\", \"right foraminal narrowing\"</p>\n<ol>\n<li><p>It's a code competition. You need to make kaggle notebook without internet access (you can use libraries and model weights as notebook input) run it on the test example, save it and commit, and after that you can submit this notebook where it evaluates your code on all test data and your score will be calculated.</p></li>\n<li><p>Yes, your code will be evaluated on unseen test data. You can try different approaches and see what works best. Join the competition, take public notebook and use it, because it's too long to doing from the scratch. But after that you need to understand how it works and what you can change to make it better. Have fun!</p></li>\n</ol>",
      "rawMarkdown": "1. Yes. But Foraminal narrowing and subarticular stenosis you have to predict for both sides - for left and right.\n\nSo you have 5 spine levels \"l1_l2\", \"l2_l3\", \"l3_l4\", \"l4-l5, \"l5-s1\" for each level you need to predict 5 conditions: \"left foraminal narrowing\", \"left subarticular stenosis\", \"canal stenosis\", \"right subarticular stenosis\", \"right foraminal narrowing\"\n\n2. It's a code competition. You need to make kaggle notebook without internet access (you can use libraries and model weights as notebook input) run it on the test example, save it and commit, and after that you can submit this notebook where it evaluates your code on all test data and your score will be calculated.\n\n3. Yes, your code will be evaluated on unseen test data. You can try different approaches and see what works best. Join the competition, take public notebook and use it, because it's too long to doing from the scratch. But after that you need to understand how it works and what you can change to make it better. Have fun!",
      "votes": null
    },
    {
      "id": "2985962",
      "postDate": "09/11/2024 07:34:33",
      "content": "<ol>\n<li><p>Yes.</p></li>\n<li><p>We need to submit a code that executes inference over a general hidden test and write the corresponding predictions in \"submission.csv\". Test samples will swap after submit click. You can find many examples on code.</p></li>\n<li><p>See 2. The approach is up to you.</p></li>\n</ol>",
      "rawMarkdown": "1. Yes.\n\n2. We need to submit a code that executes inference over a general hidden test and write the corresponding predictions in \"submission.csv\". Test samples will swap after submit click. You can find many examples on code.\n\n3. See 2. The approach is up to you.",
      "votes": null
    },
    {
      "id": "2987580",
      "postDate": "09/12/2024 18:58:35",
      "content": "<p>About the evaluation , can you explain how that is done </p>",
      "rawMarkdown": "About the evaluation , can you explain how that is done",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2985960,
      "author_name": "ambisinistra",
      "author_url": "",
      "post_date": "09/11/2024 07:33:36",
      "content": "<ol>\n<li>Yes. But Foraminal narrowing and subarticular stenosis you have to predict for both sides - for left and right.</li>\n</ol>\n<p>So you have 5 spine levels \"l1_l2\", \"l2_l3\", \"l3_l4\", \"l4-l5, \"l5-s1\" for each level you need to predict 5 conditions: \"left foraminal narrowing\", \"left subarticular stenosis\", \"canal stenosis\", \"right subarticular stenosis\", \"right foraminal narrowing\"</p>\n<ol>\n<li><p>It's a code competition. You need to make kaggle notebook without internet access (you can use libraries and model weights as notebook input) run it on the test example, save it and commit, and after that you can submit this notebook where it evaluates your code on all test data and your score will be calculated.</p></li>\n<li><p>Yes, your code will be evaluated on unseen test data. You can try different approaches and see what works best. Join the competition, take public notebook and use it, because it's too long to doing from the scratch. But after that you need to understand how it works and what you can change to make it better. Have fun!</p></li>\n</ol>",
      "votes": null,
      "replies": [
        {
          "id": 2987580,
          "author_name": "samanton",
          "author_url": "",
          "post_date": "09/12/2024 18:58:35",
          "content": "<p>About the evaluation , can you explain how that is done </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2985962,
      "author_name": "sacuscreed",
      "author_url": "",
      "post_date": "09/11/2024 07:34:33",
      "content": "<ol>\n<li><p>Yes.</p></li>\n<li><p>We need to submit a code that executes inference over a general hidden test and write the corresponding predictions in \"submission.csv\". Test samples will swap after submit click. You can find many examples on code.</p></li>\n<li><p>See 2. The approach is up to you.</p></li>\n</ol>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2985858": "Hey, I'm trying to wrap my head around this competition, but I'm having a hard time doing so since I'm new to this. Here's what I've understood so far:\n\n1. We need to build a model that classifies Foraminal Narrowing, Subarticular Stenosis, and Canal Stenosis at each spinal level.\n2. Once we’ve built the model, we either submit the model itself or just the output predictions.\n3. After submission, how will the evaluation be done? Will they run my model on their dataset and calculate the score from the results? If that's the case, do I need to adjust my model to ensure it identifies the best images where the spine is most visible?",
    "2985960": "1. Yes. But Foraminal narrowing and subarticular stenosis you have to predict for both sides - for left and right.\n\nSo you have 5 spine levels \"l1_l2\", \"l2_l3\", \"l3_l4\", \"l4-l5, \"l5-s1\" for each level you need to predict 5 conditions: \"left foraminal narrowing\", \"left subarticular stenosis\", \"canal stenosis\", \"right subarticular stenosis\", \"right foraminal narrowing\"\n\n2. It's a code competition. You need to make kaggle notebook without internet access (you can use libraries and model weights as notebook input) run it on the test example, save it and commit, and after that you can submit this notebook where it evaluates your code on all test data and your score will be calculated.\n\n3. Yes, your code will be evaluated on unseen test data. You can try different approaches and see what works best. Join the competition, take public notebook and use it, because it's too long to doing from the scratch. But after that you need to understand how it works and what you can change to make it better. Have fun!",
    "2985962": "1. Yes.\n\n2. We need to submit a code that executes inference over a general hidden test and write the corresponding predictions in \"submission.csv\". Test samples will swap after submit click. You can find many examples on code.\n\n3. See 2. The approach is up to you.",
    "2987580": "About the evaluation , can you explain how that is done"
  },
  "source": "meta"
}