{
  "id": 71726,
  "title": "so confusing about loss of validation_set",
  "url": "/competitions/quickdraw-doodle-recognition/discussion/71726",
  "author_name": "yankun",
  "post_date": "2018-11-16T02:18:39.945000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Using Resnet50, 1w images per class for testing, 2000 images per class for validation. After one epoch, the accuracy of test_set is about 70%, but for val_set, the accuracy is 30%!!!. I am sure this is not caused by overfitting, but how to explain it ???</p>",
  "messages": [
    {
      "id": 422300,
      "postDate": "2018-11-16T03:13:07.767Z",
      "content": "<p>Do you use pytorch?\nyou can check here, \n<a href=\"https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71479\">https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71479</a>\n<a href=\"https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71366\">https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71366</a></p>",
      "rawMarkdown": "Do you use pytorch?\nyou can check here, \nhttps://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71479\nhttps://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71366",
      "votes": 1,
      "replies": [
        {
          "id": 422367,
          "postDate": "2018-11-16T05:59:17.580Z",
          "content": "<p>Yes,I use pytorch! Thank you , I'll try</p>",
          "rawMarkdown": "Yes,I use pytorch! Thank you , I'll try"
        }
      ]
    },
    {
      "id": 422277,
      "postDate": "2018-11-16T02:18:39.947Z",
      "content": "<p>Using Resnet50, 1w images per class for testing, 2000 images per class for validation. After one epoch, the accuracy of test_set is about 70%, but for val_set, the accuracy is 30%!!!. I am sure this is not caused by overfitting, but how to explain it ???</p>",
      "rawMarkdown": "Using Resnet50, 1w images per class for testing, 2000 images per class for validation. After one epoch, the accuracy of test_set is about 70%, but for val_set, the accuracy is 30%!!!. I am sure this is not caused by overfitting, but how to explain it ???",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 422300,
      "author_name": "Finlay",
      "author_url": "",
      "post_date": "2018-11-16T03:13:07.767000",
      "content": "<p>Do you use pytorch?\nyou can check here, \n<a href=\"https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71479\">https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71479</a>\n<a href=\"https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71366\">https://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71366</a></p>",
      "votes": 1,
      "replies": [
        {
          "id": 422367,
          "author_name": "yankun",
          "author_url": "",
          "post_date": "2018-11-16T05:59:17.580000",
          "content": "<p>Yes,I use pytorch! Thank you , I'll try</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "422300": "Do you use pytorch?\nyou can check here, \nhttps://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71479\nhttps://www.kaggle.com/c/quickdraw-doodle-recognition/discussion/71366",
    "422277": "Using Resnet50, 1w images per class for testing, 2000 images per class for validation. After one epoch, the accuracy of test_set is about 70%, but for val_set, the accuracy is 30%!!!. I am sure this is not caused by overfitting, but how to explain it ???"
  }
}