{
  "id": 215069,
  "title": "Model overfitting",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/215069",
  "author_name": "",
  "post_date": "2021-01-28T14:59:21.788094900Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi, can anyone share what their training accuracies tend to be around for the best validation accuracies when training? When I was training, my 5 fold training accuracies tend to go to 91% while the validation accuracies stay around 89.2% - 90%? Is this normal because when submitting, my submission accuracy is only 89.4% (without TTA)?</p>",
  "messages": [
    {
      "id": "1174543",
      "postDate": "01/28/2021 14:59:21",
      "content": "<p>Hi, can anyone share what their training accuracies tend to be around for the best validation accuracies when training? When I was training, my 5 fold training accuracies tend to go to 91% while the validation accuracies stay around 89.2% - 90%? Is this normal because when submitting, my submission accuracy is only 89.4% (without TTA)?</p>",
      "rawMarkdown": "Hi, can anyone share what their training accuracies tend to be around for the best validation accuracies when training? When I was training, my 5 fold training accuracies tend to go to 91% while the validation accuracies stay around 89.2% - 90%? Is this normal because when submitting, my submission accuracy is only 89.4% (without TTA)?",
      "votes": null
    },
    {
      "id": "1174861",
      "postDate": "01/28/2021 18:54:19",
      "content": "<p>I think it is Normal.<br>\nlow validation accuracy is probably because of <br>\nhighly unbalanced Dataset.<br>\nBTW congratulations for such a great Score.</p>",
      "rawMarkdown": "I think it is Normal.\nlow validation accuracy is probably because of \nhighly unbalanced Dataset.\nBTW congratulations for such a great Score.",
      "votes": null
    },
    {
      "id": "1175255",
      "postDate": "01/29/2021 03:42:28",
      "content": "<p>Hello!</p>\n<p>There's at least one <strong><a href=\"https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/214018\" target=\"_blank\">other thread</a></strong> on overfitting. You can check it out.</p>",
      "rawMarkdown": "Hello!\n\nThere's at least one **[other thread](https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/214018)** on overfitting. You can check it out.",
      "votes": null
    },
    {
      "id": "1175290",
      "postDate": "01/29/2021 04:20:43",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/ayurveda123\" target=\"_blank\">@ayurveda123</a>, you need not to worry about  whether this model is overfitted or not. It's completely better model. </p>\n<p>In testing set they have hidden set of about 15,000 images. So, testing set accuracy was slightly changed.</p>",
      "rawMarkdown": "Hi @ayurveda123, you need not to worry about  whether this model is overfitted or not. It's completely better model. \n\nIn testing set they have hidden set of about 15,000 images. So, testing set accuracy was slightly changed.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1174861,
      "author_name": "ayannareda",
      "author_url": "",
      "post_date": "01/28/2021 18:54:19",
      "content": "<p>I think it is Normal.<br>\nlow validation accuracy is probably because of <br>\nhighly unbalanced Dataset.<br>\nBTW congratulations for such a great Score.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1175255,
      "author_name": "nickuzmenkov",
      "author_url": "",
      "post_date": "01/29/2021 03:42:28",
      "content": "<p>Hello!</p>\n<p>There's at least one <strong><a href=\"https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/214018\" target=\"_blank\">other thread</a></strong> on overfitting. You can check it out.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1175290,
      "author_name": "dhurky",
      "author_url": "",
      "post_date": "01/29/2021 04:20:43",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/ayurveda123\" target=\"_blank\">@ayurveda123</a>, you need not to worry about  whether this model is overfitted or not. It's completely better model. </p>\n<p>In testing set they have hidden set of about 15,000 images. So, testing set accuracy was slightly changed.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1174543": "Hi, can anyone share what their training accuracies tend to be around for the best validation accuracies when training? When I was training, my 5 fold training accuracies tend to go to 91% while the validation accuracies stay around 89.2% - 90%? Is this normal because when submitting, my submission accuracy is only 89.4% (without TTA)?",
    "1174861": "I think it is Normal.\nlow validation accuracy is probably because of \nhighly unbalanced Dataset.\nBTW congratulations for such a great Score.",
    "1175255": "Hello!\n\nThere's at least one **[other thread](https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/214018)** on overfitting. You can check it out.",
    "1175290": "Hi @ayurveda123, you need not to worry about  whether this model is overfitted or not. It's completely better model. \n\nIn testing set they have hidden set of about 15,000 images. So, testing set accuracy was slightly changed."
  },
  "source": "meta"
}