{
  "id": 107529,
  "title": "qwk might be unstable",
  "url": "/competitions/aptos2019-blindness-detection/discussion/107529",
  "author_name": "yshan",
  "post_date": "2019-09-05T02:25:11.447000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Improved my 7 models LB score around 0.001-0.008 by adding more  train/test time augmentations.\nMy ensemble LB score remains unchanged using the same blend weight.\nBTW I could reach 0.81 LB score by blending 10-20 public kernel submissions, but these submissions do 0 effect when I blend to my current  submission which is around 0.812.\nThis is really weird.</p>",
  "messages": [
    {
      "id": 618227,
      "postDate": "2019-09-05T02:25:11.447Z",
      "content": "<p>Improved my 7 models LB score around 0.001-0.008 by adding more  train/test time augmentations.\nMy ensemble LB score remains unchanged using the same blend weight.\nBTW I could reach 0.81 LB score by blending 10-20 public kernel submissions, but these submissions do 0 effect when I blend to my current  submission which is around 0.812.\nThis is really weird.</p>",
      "rawMarkdown": "Improved my 7 models LB score around 0.001-0.008 by adding more  train/test time augmentations.\nMy ensemble LB score remains unchanged using the same blend weight.\nBTW I could reach 0.81 LB score by blending 10-20 public kernel submissions, but these submissions do 0 effect when I blend to my current  submission which is around 0.812.\nThis is really weird.\n ",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "618227": "Improved my 7 models LB score around 0.001-0.008 by adding more  train/test time augmentations.\nMy ensemble LB score remains unchanged using the same blend weight.\nBTW I could reach 0.81 LB score by blending 10-20 public kernel submissions, but these submissions do 0 effect when I blend to my current  submission which is around 0.812.\nThis is really weird.\n "
  }
}