{
  "id": 46635,
  "title": "0.84(LB) csv file here",
  "url": "/competitions/tensorflow-speech-recognition-challenge/discussion/46635",
  "author_name": "",
  "post_date": "2017-12-31T05:49:30.666282700Z",
  "votes": -2,
  "comment_count": 8,
  "views": 0,
  "content": "<p>I used vgg11 (batch normalization, FCN, global avg pool). After trying out several models, I found that each model has different ability to categorize different words. I think this will help if you want to find the difference between different models. \nIf you have any discovery about it, you can also share here, haha.</p>",
  "messages": [
    {
      "id": "263641",
      "postDate": "12/31/2017 05:49:30",
      "content": "<p>I used vgg11 (batch normalization, FCN, global avg pool). After trying out several models, I found that each model has different ability to categorize different words. I think this will help if you want to find the difference between different models. \nIf you have any discovery about it, you can also share here, haha.</p>",
      "rawMarkdown": "I used vgg11 (batch normalization, FCN, global avg pool). After trying out several models, I found that each model has different ability to categorize different words. I think this will help if you want to find the difference between different models. \nIf you have any discovery about it, you can also share here, haha.",
      "votes": null
    },
    {
      "id": "263724",
      "postDate": "12/31/2017 15:54:51",
      "content": "<p>The code might be useful, I don't see any value to just the submission csv</p>",
      "rawMarkdown": "The code might be useful, I don't see any value to just the submission csv",
      "votes": null
    },
    {
      "id": "263745",
      "postDate": "12/31/2017 18:35:38",
      "content": "<p>I agree, In fact I see harm to those working hard with scores below 0.84 </p>",
      "rawMarkdown": "I agree, In fact I see harm to those working hard with scores below 0.84",
      "votes": null
    },
    {
      "id": "263756",
      "postDate": "12/31/2017 19:37:10",
      "content": "<p>Sharing your code in Kernels might be more helpful. Even if it is messy most people might find it more helpful than a submission.</p>",
      "rawMarkdown": "Sharing your code in Kernels might be more helpful. Even if it is messy most people might find it more helpful than a submission.",
      "votes": null
    },
    {
      "id": "263759",
      "postDate": "12/31/2017 19:46:02",
      "content": "<p>Yeah, give your code man, submission is useless.</p>",
      "rawMarkdown": "Yeah, give your code man, submission is useless.",
      "votes": null
    },
    {
      "id": "263842",
      "postDate": "01/01/2018 08:46:51",
      "content": "<p>OK, I will try to share my code in kernel later.\nI think you can find our differences, and listen to those different classifications, and finally find the shortcomings of your neural network.\nIf you find your classifier is weak in comparing long words, such as 'right' 'stop', that may mean your data-enhanced code is cutting too much during training.\nIf you find your classifier is weak in short words, such as 'up' 'off'..., that may mean your networks' receptive filed is too big to distinguish the two classes.</p>",
      "rawMarkdown": "OK, I will try to share my code in kernel later.\nI think you can find our differences, and listen to those different classifications, and finally find the shortcomings of your neural network.\nIf you find your classifier is weak in comparing long words, such as 'right' 'stop', that may mean your data-enhanced code is cutting too much during training.\nIf you find your classifier is weak in short words, such as 'up' 'off'..., that may mean your networks' receptive filed is too big to distinguish the two classes.",
      "votes": null
    },
    {
      "id": "263843",
      "postDate": "01/01/2018 08:50:28",
      "content": "<p>I think you can find our differences, and listen to those different classifications, and finally find the shortcomings of your neural network. If you find your classifier is weak in comparing long words, such as 'right' 'stop', that may mean your data-enhanced code is cutting too much during training. If you find your classifier is weak in short words, such as 'up' 'off'..., that may mean your networks' receptive filed is too big to distinguish the two classes.         Maybe it's a follow-up work. The first thing is to boost the networks' ability to fit enough data.</p>",
      "rawMarkdown": "I think you can find our differences, and listen to those different classifications, and finally find the shortcomings of your neural network. If you find your classifier is weak in comparing long words, such as 'right' 'stop', that may mean your data-enhanced code is cutting too much during training. If you find your classifier is weak in short words, such as 'up' 'off'..., that may mean your networks' receptive filed is too big to distinguish the two classes.         Maybe it's a follow-up work. The first thing is to boost the networks' ability to fit enough data.",
      "votes": null
    },
    {
      "id": "263844",
      "postDate": "01/01/2018 08:52:20",
      "content": "<p>:(. Don't worry about it. The file has been deleted. </p>",
      "rawMarkdown": ":(. Don't worry about it. The file has been deleted.",
      "votes": null
    },
    {
      "id": "263845",
      "postDate": "01/01/2018 08:55:27",
      "content": "<p>OK. I am a newbie in Kaggle. I will try to run a kernel later :D</p>",
      "rawMarkdown": "OK. I am a newbie in Kaggle. I will try to run a kernel later :D",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 263724,
      "author_name": "jfkingiii",
      "author_url": "",
      "post_date": "12/31/2017 15:54:51",
      "content": "<p>The code might be useful, I don't see any value to just the submission csv</p>",
      "votes": null,
      "replies": [
        {
          "id": 263745,
          "author_name": "ironbar",
          "author_url": "",
          "post_date": "12/31/2017 18:35:38",
          "content": "<p>I agree, In fact I see harm to those working hard with scores below 0.84 </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 263842,
          "author_name": "jihaoliu",
          "author_url": "",
          "post_date": "01/01/2018 08:46:51",
          "content": "<p>OK, I will try to share my code in kernel later.\nI think you can find our differences, and listen to those different classifications, and finally find the shortcomings of your neural network.\nIf you find your classifier is weak in comparing long words, such as 'right' 'stop', that may mean your data-enhanced code is cutting too much during training.\nIf you find your classifier is weak in short words, such as 'up' 'off'..., that may mean your networks' receptive filed is too big to distinguish the two classes.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 263844,
          "author_name": "jihaoliu",
          "author_url": "",
          "post_date": "01/01/2018 08:52:20",
          "content": "<p>:(. Don't worry about it. The file has been deleted. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 263756,
      "author_name": "CVxTz",
      "author_url": "",
      "post_date": "12/31/2017 19:37:10",
      "content": "<p>Sharing your code in Kernels might be more helpful. Even if it is messy most people might find it more helpful than a submission.</p>",
      "votes": null,
      "replies": [
        {
          "id": 263845,
          "author_name": "jihaoliu",
          "author_url": "",
          "post_date": "01/01/2018 08:55:27",
          "content": "<p>OK. I am a newbie in Kaggle. I will try to run a kernel later :D</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 263759,
      "author_name": "lugi777",
      "author_url": "",
      "post_date": "12/31/2017 19:46:02",
      "content": "<p>Yeah, give your code man, submission is useless.</p>",
      "votes": null,
      "replies": [
        {
          "id": 263843,
          "author_name": "jihaoliu",
          "author_url": "",
          "post_date": "01/01/2018 08:50:28",
          "content": "<p>I think you can find our differences, and listen to those different classifications, and finally find the shortcomings of your neural network. If you find your classifier is weak in comparing long words, such as 'right' 'stop', that may mean your data-enhanced code is cutting too much during training. If you find your classifier is weak in short words, such as 'up' 'off'..., that may mean your networks' receptive filed is too big to distinguish the two classes.         Maybe it's a follow-up work. The first thing is to boost the networks' ability to fit enough data.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "263641": "I used vgg11 (batch normalization, FCN, global avg pool). After trying out several models, I found that each model has different ability to categorize different words. I think this will help if you want to find the difference between different models. \nIf you have any discovery about it, you can also share here, haha.",
    "263724": "The code might be useful, I don't see any value to just the submission csv",
    "263745": "I agree, In fact I see harm to those working hard with scores below 0.84",
    "263756": "Sharing your code in Kernels might be more helpful. Even if it is messy most people might find it more helpful than a submission.",
    "263759": "Yeah, give your code man, submission is useless.",
    "263842": "OK, I will try to share my code in kernel later.\nI think you can find our differences, and listen to those different classifications, and finally find the shortcomings of your neural network.\nIf you find your classifier is weak in comparing long words, such as 'right' 'stop', that may mean your data-enhanced code is cutting too much during training.\nIf you find your classifier is weak in short words, such as 'up' 'off'..., that may mean your networks' receptive filed is too big to distinguish the two classes.",
    "263843": "I think you can find our differences, and listen to those different classifications, and finally find the shortcomings of your neural network. If you find your classifier is weak in comparing long words, such as 'right' 'stop', that may mean your data-enhanced code is cutting too much during training. If you find your classifier is weak in short words, such as 'up' 'off'..., that may mean your networks' receptive filed is too big to distinguish the two classes.         Maybe it's a follow-up work. The first thing is to boost the networks' ability to fit enough data.",
    "263844": ":(. Don't worry about it. The file has been deleted.",
    "263845": "OK. I am a newbie in Kaggle. I will try to run a kernel later :D"
  },
  "source": "meta"
}