{
  "id": 357386,
  "title": "Model outputs just one label for an accuracy of 72.5% (%of CE in the dataset)",
  "url": "/competitions/mayo-clinic-strip-ai/discussion/357386",
  "author_name": "joshiShivansh",
  "post_date": "2022-10-04T06:31:10.699000",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have build a model based on pretrained EfficientNet, but the model learns to output just one class (CE) and gets stuck at that. I've tried oversampling, but it hasn't worked. I also changed my LR to multiple values, but the result is the same. Can anyone suggest me what should I do?</p>",
  "messages": [
    {
      "id": 1970574,
      "postDate": "2022-10-04T06:31:10.700Z",
      "content": "<p>I have build a model based on pretrained EfficientNet, but the model learns to output just one class (CE) and gets stuck at that. I've tried oversampling, but it hasn't worked. I also changed my LR to multiple values, but the result is the same. Can anyone suggest me what should I do?</p>",
      "rawMarkdown": "I have build a model based on pretrained EfficientNet, but the model learns to output just one class (CE) and gets stuck at that. I've tried oversampling, but it hasn't worked. I also changed my LR to multiple values, but the result is the same. Can anyone suggest me what should I do?",
      "votes": 2
    },
    {
      "id": 1973000,
      "postDate": "2022-10-05T12:25:57.270Z",
      "content": "<p>A possible solution is changing your loss function, you can share what loss function you are using for further help.</p>",
      "rawMarkdown": "A possible solution is changing your loss function, you can share what loss function you are using for further help."
    },
    {
      "id": 1971588,
      "postDate": "2022-10-04T17:26:34.547Z",
      "content": "<p>Some possible solutions you can try - </p>\n<ol>\n<li>Class weights in your loss function</li>\n<li>Add Image augmentations to increase diversity</li>\n<li>Try different schedulers and observe your training loss</li>\n</ol>",
      "rawMarkdown": "Some possible solutions you can try - \n1. Class weights in your loss function\n2. Add Image augmentations to increase diversity\n3. Try different schedulers and observe your training loss"
    }
  ],
  "comments": [
    {
      "id": 1973000,
      "author_name": "Fanatic Lizard",
      "author_url": "",
      "post_date": "2022-10-05T12:25:57.270000",
      "content": "<p>A possible solution is changing your loss function, you can share what loss function you are using for further help.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1971588,
      "author_name": "Yerram Varun",
      "author_url": "",
      "post_date": "2022-10-04T17:26:34.547000",
      "content": "<p>Some possible solutions you can try - </p>\n<ol>\n<li>Class weights in your loss function</li>\n<li>Add Image augmentations to increase diversity</li>\n<li>Try different schedulers and observe your training loss</li>\n</ol>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1970574": "I have build a model based on pretrained EfficientNet, but the model learns to output just one class (CE) and gets stuck at that. I've tried oversampling, but it hasn't worked. I also changed my LR to multiple values, but the result is the same. Can anyone suggest me what should I do?",
    "1973000": "A possible solution is changing your loss function, you can share what loss function you are using for further help.",
    "1971588": "Some possible solutions you can try - \n1. Class weights in your loss function\n2. Add Image augmentations to increase diversity\n3. Try different schedulers and observe your training loss"
  }
}