{
  "id": 108000,
  "title": "please release private dataset",
  "url": "/competitions/aptos2019-blindness-detection/discussion/108000",
  "author_name": "",
  "post_date": "2019-09-08T11:45:20.115320900Z",
  "votes": 3,
  "comment_count": 1,
  "views": 0,
  "content": "<p>It could help us to see the difference of public and private data set, as @naivelamb analysis, to avoid such Inconsistent between Public/Private.</p>\n\n<p>More important, with more private data, we could test our model's generalizability on 2019 and 2015.</p>",
  "messages": [
    {
      "id": "621270",
      "postDate": "09/08/2019 11:45:20",
      "content": "<p>It could help us to see the difference of public and private data set, as @naivelamb analysis, to avoid such Inconsistent between Public/Private.</p>\n\n<p>More important, with more private data, we could test our model's generalizability on 2019 and 2015.</p>",
      "rawMarkdown": "It could help us to see the difference of public and private data set, as @naivelamb analysis, to avoid such Inconsistent between Public/Private.\n\nMore important, with more private data, we could test our model's generalizability on 2019 and 2015.",
      "votes": null
    },
    {
      "id": "621331",
      "postDate": "09/08/2019 12:46:03",
      "content": "<p>It could help us to evaluate the pratical result of our model.\nBased on local CV, the most loss point in kappa is label 3 and label 4,  because the data is very imbalance and label 3-4 is difficult for the doctors.  But is it important for doctors to provide Treatment programs?</p>\n\n<p>For the screen purpose, our models work quiet well, the recall of disease(label 1-4) is very high as binary classification problem. </p>",
      "rawMarkdown": "It could help us to evaluate the pratical result of our model.\nBased on local CV, the most loss point in kappa is label 3 and label 4,  because the data is very imbalance and label 3-4 is difficult for the doctors.  But is it important for doctors to provide Treatment programs?\n\nFor the screen purpose, our models work quiet well, the recall of disease(label 1-4) is very high as binary classification problem.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 621331,
      "author_name": "abnerzhang",
      "author_url": "",
      "post_date": "09/08/2019 12:46:03",
      "content": "<p>It could help us to evaluate the pratical result of our model.\nBased on local CV, the most loss point in kappa is label 3 and label 4,  because the data is very imbalance and label 3-4 is difficult for the doctors.  But is it important for doctors to provide Treatment programs?</p>\n\n<p>For the screen purpose, our models work quiet well, the recall of disease(label 1-4) is very high as binary classification problem. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "621270": "It could help us to see the difference of public and private data set, as @naivelamb analysis, to avoid such Inconsistent between Public/Private.\n\nMore important, with more private data, we could test our model's generalizability on 2019 and 2015.",
    "621331": "It could help us to evaluate the pratical result of our model.\nBased on local CV, the most loss point in kappa is label 3 and label 4,  because the data is very imbalance and label 3-4 is difficult for the doctors.  But is it important for doctors to provide Treatment programs?\n\nFor the screen purpose, our models work quiet well, the recall of disease(label 1-4) is very high as binary classification problem."
  },
  "source": "meta"
}