{
  "id": 91779,
  "title": "Huge difference between local cv and public LB",
  "url": "/competitions/cassava-disease/discussion/91779",
  "author_name": "",
  "post_date": "2019-05-09T01:28:39.248964900Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>With ResNet50\nLocal CV ~0.84\nLB ~ 0.33\nWith MobileNet\nLocal CV ~0.95\nLB ~0.47</p>\n\n<p>Does simple Image classification doesn't work for the problem? </p>",
  "messages": [
    {
      "id": "528975",
      "postDate": "05/09/2019 01:28:39",
      "content": "<p>With ResNet50\nLocal CV ~0.84\nLB ~ 0.33\nWith MobileNet\nLocal CV ~0.95\nLB ~0.47</p>\n\n<p>Does simple Image classification doesn't work for the problem? </p>",
      "rawMarkdown": "With ResNet50\nLocal CV ~0.84\nLB ~ 0.33\nWith MobileNet\nLocal CV ~0.95\nLB ~0.47\n\nDoes simple Image classification doesn't work for the problem?",
      "votes": null
    },
    {
      "id": "530592",
      "postDate": "05/13/2019 08:16:12",
      "content": "<p>As the dataset is imbalanced, you should carefully choose your validation set. A good one from my perspective would be the same random sample size from each category like 25% data from each category.</p>",
      "rawMarkdown": "As the dataset is imbalanced, you should carefully choose your validation set. A good one from my perspective would be the same random sample size from each category like 25% data from each category.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 530592,
      "author_name": "iskode",
      "author_url": "",
      "post_date": "05/13/2019 08:16:12",
      "content": "<p>As the dataset is imbalanced, you should carefully choose your validation set. A good one from my perspective would be the same random sample size from each category like 25% data from each category.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "528975": "With ResNet50\nLocal CV ~0.84\nLB ~ 0.33\nWith MobileNet\nLocal CV ~0.95\nLB ~0.47\n\nDoes simple Image classification doesn't work for the problem?",
    "530592": "As the dataset is imbalanced, you should carefully choose your validation set. A good one from my perspective would be the same random sample size from each category like 25% data from each category."
  },
  "source": "meta"
}