{
  "id": 229265,
  "title": "Differents accuracy",
  "url": "/competitions/digit-recognizer/discussion/229265",
  "author_name": "JeanBaptiste128",
  "post_date": "2021-03-29T10:40:32.032000",
  "votes": 0,
  "comment_count": 1,
  "views": null,
  "content": "<p>Hello,<br>\nwhen running keras I get at the minimum val_loss :<br>\nEpoch 12/1000<br>\n68/68 [==============================] - 221s 3s/step - loss: 7.7000e-05 - binary_accuracy: 1.0000 - val_loss: 0.0075 - val_binary_accuracy: 0.9979</p>\n<p>and when calculating the val_binary_accuracy by<br>\naccuracy_score(y_val,y_pred_val)<br>\nI get 0.9897</p>\n<p>Why this difference ?<br>\nThanks<br>\nJean-Baptiste</p>",
  "messages": [
    {
      "id": 1299198,
      "postDate": "2021-05-09T14:33:25.233Z",
      "content": "<p>I would say this comes from a different threshold rate between the \"binary accuracy\" and \"accuracy\" functions in keras. Maybe this blog post gives you a hint:</p>\n<p><a href=\"https://stackoverflow.com/questions/65361359/why-the-accuracy-and-binary-accuracy-in-keras-have-same-result\" target=\"_blank\">https://stackoverflow.com/questions/65361359/why-the-accuracy-and-binary-accuracy-in-keras-have-same-result</a></p>",
      "rawMarkdown": "I would say this comes from a different threshold rate between the \"binary accuracy\" and \"accuracy\" functions in keras. Maybe this blog post gives you a hint:\n\nhttps://stackoverflow.com/questions/65361359/why-the-accuracy-and-binary-accuracy-in-keras-have-same-result"
    },
    {
      "id": 1255904,
      "postDate": "2021-03-29T10:40:32.033Z",
      "content": "<p>Hello,<br>\nwhen running keras I get at the minimum val_loss :<br>\nEpoch 12/1000<br>\n68/68 [==============================] - 221s 3s/step - loss: 7.7000e-05 - binary_accuracy: 1.0000 - val_loss: 0.0075 - val_binary_accuracy: 0.9979</p>\n<p>and when calculating the val_binary_accuracy by<br>\naccuracy_score(y_val,y_pred_val)<br>\nI get 0.9897</p>\n<p>Why this difference ?<br>\nThanks<br>\nJean-Baptiste</p>",
      "rawMarkdown": "Hello,\nwhen running keras I get at the minimum val_loss :\nEpoch 12/1000\n68/68 [==============================] - 221s 3s/step - loss: 7.7000e-05 - binary_accuracy: 1.0000 - val_loss: 0.0075 - val_binary_accuracy: 0.9979\n\nand when calculating the val_binary_accuracy by\naccuracy_score(y_val,y_pred_val)\nI get 0.9897\n\nWhy this difference ?\nThanks\nJean-Baptiste"
    }
  ],
  "comments": [
    {
      "id": 1299198,
      "author_name": "Jannik Kalkmann",
      "author_url": "",
      "post_date": "2021-05-09T14:33:25.233000",
      "content": "<p>I would say this comes from a different threshold rate between the \"binary accuracy\" and \"accuracy\" functions in keras. Maybe this blog post gives you a hint:</p>\n<p><a href=\"https://stackoverflow.com/questions/65361359/why-the-accuracy-and-binary-accuracy-in-keras-have-same-result\" target=\"_blank\">https://stackoverflow.com/questions/65361359/why-the-accuracy-and-binary-accuracy-in-keras-have-same-result</a></p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1299198": "I would say this comes from a different threshold rate between the \"binary accuracy\" and \"accuracy\" functions in keras. Maybe this blog post gives you a hint:\n\nhttps://stackoverflow.com/questions/65361359/why-the-accuracy-and-binary-accuracy-in-keras-have-same-result",
    "1255904": "Hello,\nwhen running keras I get at the minimum val_loss :\nEpoch 12/1000\n68/68 [==============================] - 221s 3s/step - loss: 7.7000e-05 - binary_accuracy: 1.0000 - val_loss: 0.0075 - val_binary_accuracy: 0.9979\n\nand when calculating the val_binary_accuracy by\naccuracy_score(y_val,y_pred_val)\nI get 0.9897\n\nWhy this difference ?\nThanks\nJean-Baptiste"
  }
}