{
  "id": 124933,
  "title": "How to find a proper pair of threshold and assignment when post processing?",
  "url": "/competitions/deepfake-detection-challenge/discussion/124933",
  "author_name": "FrazierLei",
  "post_date": "2020-01-07T15:20:48.839000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Say, if I got (REAL, FAKE) = (0.3, 0.7) after softmax,  should I do some post-processing?\nI replaced all the labels that larger than 0.6 to 0.8, and others to 0.2. The threshold 0.6 and the assignment 0.8/0.2 were picked casually, and I got a decrease on LB. </p>",
  "messages": [
    {
      "id": 712782,
      "postDate": "2020-01-07T15:20:48.840Z",
      "content": "<p>Say, if I got (REAL, FAKE) = (0.3, 0.7) after softmax,  should I do some post-processing?\nI replaced all the labels that larger than 0.6 to 0.8, and others to 0.2. The threshold 0.6 and the assignment 0.8/0.2 were picked casually, and I got a decrease on LB. </p>",
      "rawMarkdown": "Say, if I got (REAL, FAKE) = (0.3, 0.7) after softmax,  should I do some post-processing?\nI replaced all the labels that larger than 0.6 to 0.8, and others to 0.2. The threshold 0.6 and the assignment 0.8/0.2 were picked casually, and I got a decrease on LB. "
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "712782": "Say, if I got (REAL, FAKE) = (0.3, 0.7) after softmax,  should I do some post-processing?\nI replaced all the labels that larger than 0.6 to 0.8, and others to 0.2. The threshold 0.6 and the assignment 0.8/0.2 were picked casually, and I got a decrease on LB. "
  }
}