{
  "id": 267439,
  "title": "Models Never Converge ??",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/267439",
  "author_name": "",
  "post_date": "2021-08-23T08:31:26.090381100Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi everyone , has anyone encountered this issue , I tried using the resnet50 pretrained with addition 4 Fully connected layers , but the model is not converging , it is struggling to minimize the loss , </p>\n<p>does anyone has encountered something similar ? or any suggestion how to get this resolved?</p>",
  "messages": [
    {
      "id": "1486826",
      "postDate": "08/23/2021 08:31:26",
      "content": "<p>Hi everyone , has anyone encountered this issue , I tried using the resnet50 pretrained with addition 4 Fully connected layers , but the model is not converging , it is struggling to minimize the loss , </p>\n<p>does anyone has encountered something similar ? or any suggestion how to get this resolved?</p>",
      "rawMarkdown": "Hi everyone , has anyone encountered this issue , I tried using the resnet50 pretrained with addition 4 Fully connected layers , but the model is not converging , it is struggling to minimize the loss , \n\ndoes anyone has encountered something similar ? or any suggestion how to get this resolved?",
      "votes": null
    },
    {
      "id": "1486886",
      "postDate": "08/23/2021 09:29:55",
      "content": "<p>Check this out <a href=\"https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/266173\" target=\"_blank\">https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/266173</a></p>",
      "rawMarkdown": "Check this out https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/266173",
      "votes": null
    },
    {
      "id": "1486927",
      "postDate": "08/23/2021 09:48:44",
      "content": "<p>Thank you for this , but i am thinking is this because we are using incorrect loss metric , what if we start to train the model with MSE loss and then observe the AUC but applying a clip on the top of the predictions ?</p>",
      "rawMarkdown": "Thank you for this , but i am thinking is this because we are using incorrect loss metric , what if we start to train the model with MSE loss and then observe the AUC but applying a clip on the top of the predictions ?",
      "votes": null
    },
    {
      "id": "1486951",
      "postDate": "08/23/2021 10:02:16",
      "content": "<p>MSE loss for classification task? I never tried it. But if you do it, do tell me how good your results are. Thanks</p>",
      "rawMarkdown": "MSE loss for classification task? I never tried it. But if you do it, do tell me how good your results are. Thanks",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1486886,
      "author_name": "mrinath",
      "author_url": "",
      "post_date": "08/23/2021 09:29:55",
      "content": "<p>Check this out <a href=\"https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/266173\" target=\"_blank\">https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/266173</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 1486927,
          "author_name": "avikrams",
          "author_url": "",
          "post_date": "08/23/2021 09:48:44",
          "content": "<p>Thank you for this , but i am thinking is this because we are using incorrect loss metric , what if we start to train the model with MSE loss and then observe the AUC but applying a clip on the top of the predictions ?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1486951,
          "author_name": "mrinath",
          "author_url": "",
          "post_date": "08/23/2021 10:02:16",
          "content": "<p>MSE loss for classification task? I never tried it. But if you do it, do tell me how good your results are. Thanks</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1486826": "Hi everyone , has anyone encountered this issue , I tried using the resnet50 pretrained with addition 4 Fully connected layers , but the model is not converging , it is struggling to minimize the loss , \n\ndoes anyone has encountered something similar ? or any suggestion how to get this resolved?",
    "1486886": "Check this out https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/266173",
    "1486927": "Thank you for this , but i am thinking is this because we are using incorrect loss metric , what if we start to train the model with MSE loss and then observe the AUC but applying a clip on the top of the predictions ?",
    "1486951": "MSE loss for classification task? I never tried it. But if you do it, do tell me how good your results are. Thanks"
  },
  "source": "meta"
}