{
  "id": 95116,
  "title": "Should we use flipping when data is enhanced?",
  "url": "/competitions/freesound-audio-tagging-2019/discussion/95116",
  "author_name": "neoguo",
  "post_date": "2019-06-10T05:58:10.244000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I saw in Kernel that many people use data flipping when data is enhanced. Indeed, I also think that the mel spectrogram data does not need to be flipped, but after I added the flip, although the training data appeared to be over-fitting. , but LB is still improving, which makes me very confused.</p>",
  "messages": [
    {
      "id": 548940,
      "postDate": "2019-06-10T05:58:10.243Z",
      "content": "<p>I saw in Kernel that many people use data flipping when data is enhanced. Indeed, I also think that the mel spectrogram data does not need to be flipped, but after I added the flip, although the training data appeared to be over-fitting. , but LB is still improving, which makes me very confused.</p>",
      "rawMarkdown": "I saw in Kernel that many people use data flipping when data is enhanced. Indeed, I also think that the mel spectrogram data does not need to be flipped, but after I added the flip, although the training data appeared to be over-fitting. , but LB is still improving, which makes me very confused."
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "548940": "I saw in Kernel that many people use data flipping when data is enhanced. Indeed, I also think that the mel spectrogram data does not need to be flipped, but after I added the flip, although the training data appeared to be over-fitting. , but LB is still improving, which makes me very confused."
  }
}