{
  "id": 102707,
  "title": "About the strategy of using the previous competition data",
  "url": "/competitions/aptos2019-blindness-detection/discussion/102707",
  "author_name": "",
  "post_date": "2019-08-04T06:28:18.529025600Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I found the most popular strategy of using the previous competition data is:\nIn stage 1, use the previous competition data as training data and at the same time using the new data as validation\nIn stage 2, training as normal on new data\nfollow the idea i even got worse LB, my question is what about the class balance of data, have you ever balanced the new data(in stage1 and stage2), and how about the old data, balance or lot?</p>",
  "messages": [
    {
      "id": "591714",
      "postDate": "08/04/2019 06:28:18",
      "content": "<p>I found the most popular strategy of using the previous competition data is:\nIn stage 1, use the previous competition data as training data and at the same time using the new data as validation\nIn stage 2, training as normal on new data\nfollow the idea i even got worse LB, my question is what about the class balance of data, have you ever balanced the new data(in stage1 and stage2), and how about the old data, balance or lot?</p>",
      "rawMarkdown": "I found the most popular strategy of using the previous competition data is:\nIn stage 1, use the previous competition data as training data and at the same time using the new data as validation\nIn stage 2, training as normal on new data\nfollow the idea i even got worse LB, my question is what about the class balance of data, have you ever balanced the new data(in stage1 and stage2), and how about the old data, balance or lot?",
      "votes": null
    },
    {
      "id": "592271",
      "postDate": "08/05/2019 05:25:12",
      "content": "<p>Having the same issue.</p>",
      "rawMarkdown": "Having the same issue.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 592271,
      "author_name": "dreimd",
      "author_url": "",
      "post_date": "08/05/2019 05:25:12",
      "content": "<p>Having the same issue.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "591714": "I found the most popular strategy of using the previous competition data is:\nIn stage 1, use the previous competition data as training data and at the same time using the new data as validation\nIn stage 2, training as normal on new data\nfollow the idea i even got worse LB, my question is what about the class balance of data, have you ever balanced the new data(in stage1 and stage2), and how about the old data, balance or lot?",
    "592271": "Having the same issue."
  },
  "source": "meta"
}