{
  "id": 48546,
  "title": "VGG19-BN scoring 0.89839, others >0.90",
  "url": "/competitions/tensorflow-speech-recognition-challenge/writeups/but-vgg19-bn-scoring-0-89839-others-0-90",
  "author_name": "",
  "post_date": "2018-01-29T15:16:10.229931600Z",
  "votes": 6,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Some people had really good results with mixup: <a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47730\">https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47730</a></p>\n\n<p>So we decided to implement a naive mixup mixing the mel-spectograms: <a href=\"https://github.com/tugstugi/pytorch-speech-commands\">https://github.com/tugstugi/pytorch-speech-commands</a></p>\n\n<p>With this simple mixup, even VGG could get 0.89839. Other models could reach at least 0.90%.</p>",
  "messages": [
    {
      "id": "275600",
      "postDate": "01/29/2018 15:16:10",
      "content": "<p>Some people had really good results with mixup: <a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47730\">https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47730</a></p>\n\n<p>So we decided to implement a naive mixup mixing the mel-spectograms: <a href=\"https://github.com/tugstugi/pytorch-speech-commands\">https://github.com/tugstugi/pytorch-speech-commands</a></p>\n\n<p>With this simple mixup, even VGG could get 0.89839. Other models could reach at least 0.90%.</p>",
      "rawMarkdown": "Some people had really good results with mixup: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47730\n\nSo we decided to implement a naive mixup mixing the mel-spectograms: https://github.com/tugstugi/pytorch-speech-commands\n\nWith this simple mixup, even VGG could get 0.89839. Other models could reach at least 0.90%.",
      "votes": null
    },
    {
      "id": "276404",
      "postDate": "01/31/2018 12:28:41",
      "content": "<p>Thank you for sharing your codes.</p>",
      "rawMarkdown": "Thank you for sharing your codes.",
      "votes": null
    },
    {
      "id": "276967",
      "postDate": "02/01/2018 23:04:55",
      "content": "<p>Hey Tugi - gonna check this out and run it!  Looks good.  Best Wishes - Humphrey from Palo Alto</p>",
      "rawMarkdown": "Hey Tugi - gonna check this out and run it!  Looks good.  Best Wishes - Humphrey from Palo Alto",
      "votes": null
    },
    {
      "id": "312208",
      "postDate": "04/11/2018 11:38:25",
      "content": "<p>Impressive results!\nDid you use cross validation? If so, are these scores the result of a single round or average of multiple folds?\n Thanks.</p>",
      "rawMarkdown": "Impressive results!\nDid you use cross validation? If so, are these scores the result of a single round or average of multiple folds?\n Thanks.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 276404,
      "author_name": "yatzhash",
      "author_url": "",
      "post_date": "01/31/2018 12:28:41",
      "content": "<p>Thank you for sharing your codes.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 276967,
      "author_name": "dolphindecode",
      "author_url": "",
      "post_date": "02/01/2018 23:04:55",
      "content": "<p>Hey Tugi - gonna check this out and run it!  Looks good.  Best Wishes - Humphrey from Palo Alto</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 312208,
      "author_name": "shijing",
      "author_url": "",
      "post_date": "04/11/2018 11:38:25",
      "content": "<p>Impressive results!\nDid you use cross validation? If so, are these scores the result of a single round or average of multiple folds?\n Thanks.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "275600": "Some people had really good results with mixup: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/47730\n\nSo we decided to implement a naive mixup mixing the mel-spectograms: https://github.com/tugstugi/pytorch-speech-commands\n\nWith this simple mixup, even VGG could get 0.89839. Other models could reach at least 0.90%.",
    "276404": "Thank you for sharing your codes.",
    "276967": "Hey Tugi - gonna check this out and run it!  Looks good.  Best Wishes - Humphrey from Palo Alto",
    "312208": "Impressive results!\nDid you use cross validation? If so, are these scores the result of a single round or average of multiple folds?\n Thanks."
  },
  "source": "meta"
}