{
  "id": 251412,
  "title": "How to augment data?",
  "url": "/competitions/seti-breakthrough-listen/discussion/251412",
  "author_name": "Dylan Liu",
  "post_date": "2021-07-07T06:49:58.291000",
  "votes": 0,
  "comment_count": 2,
  "views": 0,
  "content": "<p>If we treat the training data as images, then it's easy to augment. The problem is that the data is kind of signal sequences (like audio), so the augmentation for images will likely not appear in the data.</p>",
  "messages": [
    {
      "id": 1379713,
      "postDate": "2021-07-07T14:45:47.940Z",
      "content": "<p>I've seen lots of people use vertical and horizontal flipping, and also MixUp. I'm also using some gaussian noise and dropout.</p>\n<p>I'd also recommend having a look at this kernel: <a href=\"https://www.kaggle.com/shionhonda/search-for-effective-data-augmentation\" target=\"_blank\">https://www.kaggle.com/shionhonda/search-for-effective-data-augmentation</a></p>",
      "rawMarkdown": "I've seen lots of people use vertical and horizontal flipping, and also MixUp. I'm also using some gaussian noise and dropout.\n\nI'd also recommend having a look at this kernel: https://www.kaggle.com/shionhonda/search-for-effective-data-augmentation",
      "votes": 4,
      "replies": [
        {
          "id": 1380097,
          "postDate": "2021-07-07T19:24:08.003Z",
          "content": "<p>I'm also using gaussian noise.</p>",
          "rawMarkdown": "I'm also using gaussian noise.",
          "votes": 1
        }
      ]
    },
    {
      "id": 1379192,
      "postDate": "2021-07-07T06:49:58.290Z",
      "content": "<p>If we treat the training data as images, then it's easy to augment. The problem is that the data is kind of signal sequences (like audio), so the augmentation for images will likely not appear in the data.</p>",
      "rawMarkdown": "If we treat the training data as images, then it's easy to augment. The problem is that the data is kind of signal sequences (like audio), so the augmentation for images will likely not appear in the data."
    }
  ],
  "comments": [
    {
      "id": 1379713,
      "author_name": "Nathan Smith",
      "author_url": "",
      "post_date": "2021-07-07T14:45:47.940000",
      "content": "<p>I've seen lots of people use vertical and horizontal flipping, and also MixUp. I'm also using some gaussian noise and dropout.</p>\n<p>I'd also recommend having a look at this kernel: <a href=\"https://www.kaggle.com/shionhonda/search-for-effective-data-augmentation\" target=\"_blank\">https://www.kaggle.com/shionhonda/search-for-effective-data-augmentation</a></p>",
      "votes": 4,
      "replies": [
        {
          "id": 1380097,
          "author_name": "عثمان",
          "author_url": "",
          "post_date": "2021-07-07T19:24:08.003000",
          "content": "<p>I'm also using gaussian noise.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1379713": "I've seen lots of people use vertical and horizontal flipping, and also MixUp. I'm also using some gaussian noise and dropout.\n\nI'd also recommend having a look at this kernel: https://www.kaggle.com/shionhonda/search-for-effective-data-augmentation",
    "1379192": "If we treat the training data as images, then it's easy to augment. The problem is that the data is kind of signal sequences (like audio), so the augmentation for images will likely not appear in the data."
  }
}