{
  "id": 234204,
  "title": "Data cleaning",
  "url": "/competitions/birdclef-2021/discussion/234204",
  "author_name": "",
  "post_date": "2021-04-23T06:37:55.589551400Z",
  "votes": 5,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Аfter numerous experiments, I came to the conclusion that manual cleaning of the data spectrograms is the best way. For clarity, I find the average for each spectral band in the spectrogram and then subtract the median of the spectrum from each frame for better visualisation, save this, look at them and clear non-informative spectograms. This greatly improves my result. But it takes a lot of time… If someone knows how to automate such a process or similar, let me know if possible) And what approaches have you tried?</p>",
  "messages": [
    {
      "id": "1281634",
      "postDate": "04/23/2021 06:37:55",
      "content": "<p>Аfter numerous experiments, I came to the conclusion that manual cleaning of the data spectrograms is the best way. For clarity, I find the average for each spectral band in the spectrogram and then subtract the median of the spectrum from each frame for better visualisation, save this, look at them and clear non-informative spectograms. This greatly improves my result. But it takes a lot of time… If someone knows how to automate such a process or similar, let me know if possible) And what approaches have you tried?</p>",
      "rawMarkdown": "Аfter numerous experiments, I came to the conclusion that manual cleaning of the data spectrograms is the best way. For clarity, I find the average for each spectral band in the spectrogram and then subtract the median of the spectrum from each frame for better visualisation, save this, look at them and clear non-informative spectograms. This greatly improves my result. But it takes a lot of time... If someone knows how to automate such a process or similar, let me know if possible) And what approaches have you tried?",
      "votes": null
    },
    {
      "id": "1281700",
      "postDate": "04/23/2021 07:46:17",
      "content": "<blockquote>\n  <p>This greatly improves my result. </p>\n</blockquote>\n<p>What do you mean by this?  You get a better LB score?</p>",
      "rawMarkdown": "> This greatly improves my result. \n\nWhat do you mean by this?  You get a better LB score?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1281700,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "04/23/2021 07:46:17",
      "content": "<blockquote>\n  <p>This greatly improves my result. </p>\n</blockquote>\n<p>What do you mean by this?  You get a better LB score?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1281634": "Аfter numerous experiments, I came to the conclusion that manual cleaning of the data spectrograms is the best way. For clarity, I find the average for each spectral band in the spectrogram and then subtract the median of the spectrum from each frame for better visualisation, save this, look at them and clear non-informative spectograms. This greatly improves my result. But it takes a lot of time... If someone knows how to automate such a process or similar, let me know if possible) And what approaches have you tried?",
    "1281700": "> This greatly improves my result. \n\nWhat do you mean by this?  You get a better LB score?"
  },
  "source": "meta"
}