{
  "id": 11445,
  "title": "has anyone tried sparse filtering?",
  "url": "/competitions/datasciencebowl/discussion/11445",
  "author_name": "",
  "post_date": "2015-01-07T07:28:54.937Z",
  "votes": 1,
  "comment_count": 1,
  "views": 1226,
  "content": "<p>I noticed sparse filtering was used in a prior Kaggle competition: http://fastml.com/deep-learning-made-easy/<br>https://github.com/zygmuntz/kaggle-blackbox</p>\n<p>Has anyone tried using it for this competition?&nbsp;</p>",
  "messages": [
    {
      "id": "62389",
      "postDate": "01/07/2015 07:28:54",
      "content": "<p>I noticed sparse filtering was used in a prior Kaggle competition: http://fastml.com/deep-learning-made-easy/<br>https://github.com/zygmuntz/kaggle-blackbox</p>\n<p>Has anyone tried using it for this competition?&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "62473",
      "postDate": "01/08/2015 23:56:20",
      "content": "<p>Hey Mike...</p>\n<p>This is a good link, thanks for sharing.</p>\n<p>Is sparse filtering similar to sparse autoencoders? &nbsp;It looks like it is at first glance.</p>\n<p>I'm using RBMs to try and accomplish the same thing. &nbsp;See the really good tutorial here: &nbsp;http://deeplearning.net/tutorial/rbm.html</p>\n<p>The papers&nbsp;I've read indicate using dropout during pretraining with RBMs develops better features, so that's what I'm trying to do. &nbsp;Not great success yet though, so it's either that my network isn't big/deep enough ([10k, 10k, 10k] is the biggest I can fit in GPU RAM right now)... &nbsp;or I've messed up my implementation of dropout... &nbsp;or both. &nbsp;It seemed to work on MNIST dataset, so I'm hoping my dropout implementation is reasonable, but we'll see.</p>\n<p>Have you had any success yet with this sparse filtering?</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 62473,
      "author_name": "kevinhinson",
      "author_url": "",
      "post_date": "01/08/2015 23:56:20",
      "content": "<p>Hey Mike...</p>\n<p>This is a good link, thanks for sharing.</p>\n<p>Is sparse filtering similar to sparse autoencoders? &nbsp;It looks like it is at first glance.</p>\n<p>I'm using RBMs to try and accomplish the same thing. &nbsp;See the really good tutorial here: &nbsp;http://deeplearning.net/tutorial/rbm.html</p>\n<p>The papers&nbsp;I've read indicate using dropout during pretraining with RBMs develops better features, so that's what I'm trying to do. &nbsp;Not great success yet though, so it's either that my network isn't big/deep enough ([10k, 10k, 10k] is the biggest I can fit in GPU RAM right now)... &nbsp;or I've messed up my implementation of dropout... &nbsp;or both. &nbsp;It seemed to work on MNIST dataset, so I'm hoping my dropout implementation is reasonable, but we'll see.</p>\n<p>Have you had any success yet with this sparse filtering?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "62389": "",
    "62473": ""
  },
  "source": "meta"
}