{
  "id": 218862,
  "title": "Why BiTemperedLogisticLoss isn't working?",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/218862",
  "author_name": "",
  "post_date": "2021-02-12T10:46:57.745605800Z",
  "votes": 1,
  "comment_count": 7,
  "views": 0,
  "content": "<p>I'm new in deep learning and kaggle. I'm using EfficientNet and I tried to use BiTemperedLogisticLoss but the loss and accuracy didn't improve at all.(loss:21.1229, accuracy:0.1947)<br>\nI don't know what is wrong. Can anyone know what is the reason?</p>\n<p>Thank you.</p>",
  "messages": [
    {
      "id": "1197701",
      "postDate": "02/12/2021 10:46:57",
      "content": "<p>I'm new in deep learning and kaggle. I'm using EfficientNet and I tried to use BiTemperedLogisticLoss but the loss and accuracy didn't improve at all.(loss:21.1229, accuracy:0.1947)<br>\nI don't know what is wrong. Can anyone know what is the reason?</p>\n<p>Thank you.</p>",
      "rawMarkdown": "I'm new in deep learning and kaggle. I'm using EfficientNet and I tried to use BiTemperedLogisticLoss but the loss and accuracy didn't improve at all.(loss:21.1229, accuracy:0.1947)\nI don't know what is wrong. Can anyone know what is the reason?\n\nThank you.",
      "votes": null
    },
    {
      "id": "1198003",
      "postDate": "02/12/2021 16:05:06",
      "content": "<p>This happened to me too. I think the implementation I used had some error.<br>\nI am using Cross entropy now.</p>",
      "rawMarkdown": "This happened to me too. I think the implementation I used had some error.\nI am using Cross entropy now.",
      "votes": null
    },
    {
      "id": "1201021",
      "postDate": "02/15/2021 05:39:37",
      "content": "<p>Bc it's just a loss function. It might give a boost or not, but it's not game changing.</p>",
      "rawMarkdown": "Bc it's just a loss function. It might give a boost or not, but it's not game changing.",
      "votes": null
    },
    {
      "id": "1204403",
      "postDate": "02/16/2021 05:50:55",
      "content": "<p>You can try adjusting values of t1 and t2. Begin with narrow change to see the impact on CV and LB.</p>",
      "rawMarkdown": "You can try adjusting values of t1 and t2. Begin with narrow change to see the impact on CV and LB.",
      "votes": null
    },
    {
      "id": "1204432",
      "postDate": "02/16/2021 06:17:30",
      "content": "<p>I made some changes in my code and I'm using sparse cross entropy now.<br>\nThank you for your comment!</p>",
      "rawMarkdown": "I made some changes in my code and I'm using sparse cross entropy now.\nThank you for your comment!",
      "votes": null
    },
    {
      "id": "1204442",
      "postDate": "02/16/2021 06:25:48",
      "content": "<p>Thank you for shearing your idea!<br>\nI'll spend time in other things.</p>",
      "rawMarkdown": "Thank you for shearing your idea!\nI'll spend time in other things.",
      "votes": null
    },
    {
      "id": "1204447",
      "postDate": "02/16/2021 06:31:04",
      "content": "<p>I tried it but CategoricalCrossentropy worked better.<br>\nThank you for your advise!</p>",
      "rawMarkdown": "I tried it but CategoricalCrossentropy worked better.\nThank you for your advise!",
      "votes": null
    },
    {
      "id": "1208567",
      "postDate": "02/18/2021 10:43:46",
      "content": "<p>Bitempered loss has 2 parameters called t1 and t2. They have limited value range,<br>\n0&lt; t1 &lt;= 1, 1 &lt;= t2     did you try?<br>\nOr you can try other optimizer and learning rate</p>",
      "rawMarkdown": "Bitempered loss has 2 parameters called t1 and t2. They have limited value range,\n0< t1 <= 1, 1 <= t2     did you try?\nOr you can try other optimizer and learning rate",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1198003,
      "author_name": "mohneesh7",
      "author_url": "",
      "post_date": "02/12/2021 16:05:06",
      "content": "<p>This happened to me too. I think the implementation I used had some error.<br>\nI am using Cross entropy now.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1204432,
          "author_name": "ryujiaoki",
          "author_url": "",
          "post_date": "02/16/2021 06:17:30",
          "content": "<p>I made some changes in my code and I'm using sparse cross entropy now.<br>\nThank you for your comment!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1201021,
      "author_name": "underwearfitting",
      "author_url": "",
      "post_date": "02/15/2021 05:39:37",
      "content": "<p>Bc it's just a loss function. It might give a boost or not, but it's not game changing.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1204442,
          "author_name": "ryujiaoki",
          "author_url": "",
          "post_date": "02/16/2021 06:25:48",
          "content": "<p>Thank you for shearing your idea!<br>\nI'll spend time in other things.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1204403,
      "author_name": "vickygoyal",
      "author_url": "",
      "post_date": "02/16/2021 05:50:55",
      "content": "<p>You can try adjusting values of t1 and t2. Begin with narrow change to see the impact on CV and LB.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1204447,
          "author_name": "ryujiaoki",
          "author_url": "",
          "post_date": "02/16/2021 06:31:04",
          "content": "<p>I tried it but CategoricalCrossentropy worked better.<br>\nThank you for your advise!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1208567,
      "author_name": "yoshito",
      "author_url": "",
      "post_date": "02/18/2021 10:43:46",
      "content": "<p>Bitempered loss has 2 parameters called t1 and t2. They have limited value range,<br>\n0&lt; t1 &lt;= 1, 1 &lt;= t2     did you try?<br>\nOr you can try other optimizer and learning rate</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1197701": "I'm new in deep learning and kaggle. I'm using EfficientNet and I tried to use BiTemperedLogisticLoss but the loss and accuracy didn't improve at all.(loss:21.1229, accuracy:0.1947)\nI don't know what is wrong. Can anyone know what is the reason?\n\nThank you.",
    "1198003": "This happened to me too. I think the implementation I used had some error.\nI am using Cross entropy now.",
    "1201021": "Bc it's just a loss function. It might give a boost or not, but it's not game changing.",
    "1204403": "You can try adjusting values of t1 and t2. Begin with narrow change to see the impact on CV and LB.",
    "1204432": "I made some changes in my code and I'm using sparse cross entropy now.\nThank you for your comment!",
    "1204442": "Thank you for shearing your idea!\nI'll spend time in other things.",
    "1204447": "I tried it but CategoricalCrossentropy worked better.\nThank you for your advise!",
    "1208567": "Bitempered loss has 2 parameters called t1 and t2. They have limited value range,\n0< t1 <= 1, 1 <= t2     did you try?\nOr you can try other optimizer and learning rate"
  },
  "source": "meta"
}