{
  "id": 43685,
  "title": "The big PyTorch Thread",
  "url": "/competitions/tensorflow-speech-recognition-challenge/discussion/43685",
  "author_name": "",
  "post_date": "2017-11-17T21:20:35.182040Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Dear All,\nAs a PyTorch advocate, I am trying to do my best to have a fully working PyTorch starter notebook. \n<a href=\"https://www.kaggle.com/solomonk/pytorch-speech-recognition-challenge-wip\">https://www.kaggle.com/solomonk/pytorch-speech-recognition-challenge-wip</a></p>\n\n<p>As always, the main issues with PyTorch (at least for me ...) are the custom data loaders. I used the script written here: <a href=\"https://www.kaggle.com/timolee/audio-data-conversion-to-images-eda\">https://www.kaggle.com/timolee/audio-data-conversion-to-images-eda</a> to convert the audio into Images for my custom data loader. </p>\n\n<p>This is the first time I write PyTorch for a multi-class classification problem, though I wrote numerous notebooks for Binary classification using PyTorch, I seem to have a problem with the loss function. Hopefully to be resolved soon. </p>\n\n<p>If you are a PyTorcher, feel free to clone my code and share here your results. </p>\n\n<p>Best, \nQuantscientist, </p>",
  "messages": [
    {
      "id": "245243",
      "postDate": "11/17/2017 21:20:35",
      "content": "<p>Dear All,\nAs a PyTorch advocate, I am trying to do my best to have a fully working PyTorch starter notebook. \n<a href=\"https://www.kaggle.com/solomonk/pytorch-speech-recognition-challenge-wip\">https://www.kaggle.com/solomonk/pytorch-speech-recognition-challenge-wip</a></p>\n\n<p>As always, the main issues with PyTorch (at least for me ...) are the custom data loaders. I used the script written here: <a href=\"https://www.kaggle.com/timolee/audio-data-conversion-to-images-eda\">https://www.kaggle.com/timolee/audio-data-conversion-to-images-eda</a> to convert the audio into Images for my custom data loader. </p>\n\n<p>This is the first time I write PyTorch for a multi-class classification problem, though I wrote numerous notebooks for Binary classification using PyTorch, I seem to have a problem with the loss function. Hopefully to be resolved soon. </p>\n\n<p>If you are a PyTorcher, feel free to clone my code and share here your results. </p>\n\n<p>Best, \nQuantscientist, </p>",
      "rawMarkdown": "Dear All,\nAs a PyTorch advocate, I am trying to do my best to have a fully working PyTorch starter notebook. \nhttps://www.kaggle.com/solomonk/pytorch-speech-recognition-challenge-wip\n\nAs always, the main issues with PyTorch (at least for me ...) are the custom data loaders. I used the script written here: https://www.kaggle.com/timolee/audio-data-conversion-to-images-eda to convert the audio into Images for my custom data loader. \n\nThis is the first time I write PyTorch for a multi-class classification problem, though I wrote numerous notebooks for Binary classification using PyTorch, I seem to have a problem with the loss function. Hopefully to be resolved soon. \n\nIf you are a PyTorcher, feel free to clone my code and share here your results. \n\nBest, \nQuantscientist,",
      "votes": null
    },
    {
      "id": "245283",
      "postDate": "11/17/2017 23:30:16",
      "content": "<p>Do we suppose to use tensorflow ?</p>",
      "rawMarkdown": "Do we suppose to use tensorflow ?",
      "votes": null
    },
    {
      "id": "245305",
      "postDate": "11/18/2017 01:56:23",
      "content": "<p>There's no requirement to use TensorFlow at all for the main contest. I'm a big fan of PyTorch myself, and I wouldn't be surprised to see other frameworks like Kaldi that are specialized for speech recognition proving useful.</p>\n\n<p>The special prize does require a TensorFlow model, because I'm keen to open source a solution that will work fast and easily on the Raspberry Pi, and restricting to a standard platform makes measuring those properties easier. As we've seen with the TF speech tutorial though, I would expect any entry could be ported to other frameworks in addition.</p>",
      "rawMarkdown": "There's no requirement to use TensorFlow at all for the main contest. I'm a big fan of PyTorch myself, and I wouldn't be surprised to see other frameworks like Kaldi that are specialized for speech recognition proving useful.\n\nThe special prize does require a TensorFlow model, because I'm keen to open source a solution that will work fast and easily on the Raspberry Pi, and restricting to a standard platform makes measuring those properties easier. As we've seen with the TF speech tutorial though, I would expect any entry could be ported to other frameworks in addition.",
      "votes": null
    },
    {
      "id": "245311",
      "postDate": "11/18/2017 02:40:09",
      "content": "<p>@wudi  If you need help with PyTorch let me know. </p>",
      "rawMarkdown": "wudi  If you need help with PyTorch let me know.",
      "votes": null
    },
    {
      "id": "245394",
      "postDate": "11/18/2017 09:55:07",
      "content": "<p><strong>Update 18/11/2107:</strong>\n- PyTorch Reference implementations provided: <a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624\">https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624</a> by \n- I rendered my notebook here on github: <a href=\"https://github.com/QuantScientist/Deep-Learning-Boot-Camp/blob/master/Kaggle-PyTorch/tf/PyTorch%20Speech%20Recognition%20Challenge%20Starter.ipynb\">https://github.com/QuantScientist/Deep-Learning-Boot-Camp/blob/master/Kaggle-PyTorch/tf/PyTorch%20Speech%20Recognition%20Challenge%20Starter.ipynb</a></p>",
      "rawMarkdown": "**Update 18/11/2107:**\n- PyTorch Reference implementations provided: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624 by \n- I rendered my notebook here on github: https://github.com/QuantScientist/Deep-Learning-Boot-Camp/blob/master/Kaggle-PyTorch/tf/PyTorch%20Speech%20Recognition%20Challenge%20Starter.ipynb",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 245283,
      "author_name": "woodywang",
      "author_url": "",
      "post_date": "11/17/2017 23:30:16",
      "content": "<p>Do we suppose to use tensorflow ?</p>",
      "votes": null,
      "replies": [
        {
          "id": 245305,
          "author_name": "petewarden",
          "author_url": "",
          "post_date": "11/18/2017 01:56:23",
          "content": "<p>There's no requirement to use TensorFlow at all for the main contest. I'm a big fan of PyTorch myself, and I wouldn't be surprised to see other frameworks like Kaldi that are specialized for speech recognition proving useful.</p>\n\n<p>The special prize does require a TensorFlow model, because I'm keen to open source a solution that will work fast and easily on the Raspberry Pi, and restricting to a standard platform makes measuring those properties easier. As we've seen with the TF speech tutorial though, I would expect any entry could be ported to other frameworks in addition.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 245311,
          "author_name": "solomonk",
          "author_url": "",
          "post_date": "11/18/2017 02:40:09",
          "content": "<p>@wudi  If you need help with PyTorch let me know. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 245394,
      "author_name": "solomonk",
      "author_url": "",
      "post_date": "11/18/2017 09:55:07",
      "content": "<p><strong>Update 18/11/2107:</strong>\n- PyTorch Reference implementations provided: <a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624\">https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624</a> by \n- I rendered my notebook here on github: <a href=\"https://github.com/QuantScientist/Deep-Learning-Boot-Camp/blob/master/Kaggle-PyTorch/tf/PyTorch%20Speech%20Recognition%20Challenge%20Starter.ipynb\">https://github.com/QuantScientist/Deep-Learning-Boot-Camp/blob/master/Kaggle-PyTorch/tf/PyTorch%20Speech%20Recognition%20Challenge%20Starter.ipynb</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "245243": "Dear All,\nAs a PyTorch advocate, I am trying to do my best to have a fully working PyTorch starter notebook. \nhttps://www.kaggle.com/solomonk/pytorch-speech-recognition-challenge-wip\n\nAs always, the main issues with PyTorch (at least for me ...) are the custom data loaders. I used the script written here: https://www.kaggle.com/timolee/audio-data-conversion-to-images-eda to convert the audio into Images for my custom data loader. \n\nThis is the first time I write PyTorch for a multi-class classification problem, though I wrote numerous notebooks for Binary classification using PyTorch, I seem to have a problem with the loss function. Hopefully to be resolved soon. \n\nIf you are a PyTorcher, feel free to clone my code and share here your results. \n\nBest, \nQuantscientist,",
    "245283": "Do we suppose to use tensorflow ?",
    "245305": "There's no requirement to use TensorFlow at all for the main contest. I'm a big fan of PyTorch myself, and I wouldn't be surprised to see other frameworks like Kaldi that are specialized for speech recognition proving useful.\n\nThe special prize does require a TensorFlow model, because I'm keen to open source a solution that will work fast and easily on the Raspberry Pi, and restricting to a standard platform makes measuring those properties easier. As we've seen with the TF speech tutorial though, I would expect any entry could be ported to other frameworks in addition.",
    "245311": "wudi  If you need help with PyTorch let me know.",
    "245394": "**Update 18/11/2107:**\n- PyTorch Reference implementations provided: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624 by \n- I rendered my notebook here on github: https://github.com/QuantScientist/Deep-Learning-Boot-Camp/blob/master/Kaggle-PyTorch/tf/PyTorch%20Speech%20Recognition%20Challenge%20Starter.ipynb"
  },
  "source": "meta"
}