{
  "id": 12839,
  "title": "Curiousity Survey. Add/upvote your toolkit.",
  "url": "/competitions/datasciencebowl/discussion/12839",
  "author_name": "",
  "post_date": "2015-03-16T21:43:39.010Z",
  "votes": 3,
  "comment_count": 37,
  "views": 6169,
  "content": "<p>Caffe</p>",
  "messages": [
    {
      "id": "66526",
      "postDate": "03/16/2015 21:43:39",
      "content": "<p>Caffe</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66527",
      "postDate": "03/16/2015 21:44:18",
      "content": "<p>NoLearn/Lasagne</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66528",
      "postDate": "03/16/2015 21:44:56",
      "content": "<p>Lasagne/nolearn</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66531",
      "postDate": "03/16/2015 21:49:26",
      "content": "<p>Caffe</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66535",
      "postDate": "03/16/2015 21:54:26",
      "content": "<p>Caffe</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66536",
      "postDate": "03/16/2015 21:54:36",
      "content": "<p>Lasagne/nolearn</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66537",
      "postDate": "03/16/2015 21:59:34",
      "content": "<p>started with Pylearn2, finished with caffe</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66539",
      "postDate": "03/16/2015 22:01:07",
      "content": "<p>I think he means upvote ... ok anyway Caffe, indeed I don't like it... took half day to install, and one day to find out how to make predictions.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66540",
      "postDate": "03/16/2015 22:03:26",
      "content": "<p>Torch</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66541",
      "postDate": "03/16/2015 22:06:40",
      "content": "<p>[quote=Steven Du;66539]</p>\n<p>I think he means upvote ... ok anyway Caffe, indeed I don't like it... took half day to install, and one day to find out how to make predictions.</p>\n<p>[/quote]</p>\n<p>LOL . . . yeah.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66543",
      "postDate": "03/16/2015 22:10:08",
      "content": "<p>Lasagne/Nolearn is painless to install and very versatile&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66561",
      "postDate": "03/16/2015 23:05:07",
      "content": "<p>Plain Theano</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66617",
      "postDate": "03/17/2015 03:10:56",
      "content": "<p>theano/sklearn</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66619",
      "postDate": "03/17/2015 03:18:38",
      "content": "<p>Torch</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66621",
      "postDate": "03/17/2015 03:20:14",
      "content": "<p>Caffe</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66627",
      "postDate": "03/17/2015 03:53:05",
      "content": "<p>Cxxnet + python for image augmentation, looking forward to the update.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66634",
      "postDate": "03/17/2015 04:58:12",
      "content": "<p>NoLearn/lasagne</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66647",
      "postDate": "03/17/2015 06:37:54",
      "content": "<p>Caffe...</p>\n<p>but installation and documentation is a pain in the a**. Did not find a way for real time training data augmentation except by brute hacking the cpp-code.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66649",
      "postDate": "03/17/2015 06:44:53",
      "content": "<p>Lasagne</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66652",
      "postDate": "03/17/2015 07:09:19",
      "content": "<p>Lasagne / Nolearn</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66658",
      "postDate": "03/17/2015 07:53:28",
      "content": "<p>CXXNET</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66661",
      "postDate": "03/17/2015 08:13:22",
      "content": "<p>mainly Caffe and a little bit of matconvnet</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66671",
      "postDate": "03/17/2015 09:27:29",
      "content": "<p>Caffe, but as I saw the winning solution, I am thinking about Torch or wrapper.&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66672",
      "postDate": "03/17/2015 09:35:16",
      "content": "<p>Caffe and lasagne+nolearn</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66693",
      "postDate": "03/17/2015 13:01:21",
      "content": "<p>cuda_convnet</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66714",
      "postDate": "03/17/2015 14:50:09",
      "content": "<p>Pylearn2</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66724",
      "postDate": "03/17/2015 15:45:34",
      "content": "<p>Torch, then CXXNET</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66728",
      "postDate": "03/17/2015 15:54:05",
      "content": "<p>Spatially Sparse ConvNets&nbsp;</p>\n<p>http://www2.warwick.ac.uk/fac/sci/statistics/staff/academic-research/graham/</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66729",
      "postDate": "03/17/2015 15:56:56",
      "content": "<p>@Ryan Munion, are you going to share yours solution with Prof. Graham as well?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66730",
      "postDate": "03/17/2015 15:57:13",
      "content": "<p>[quote=Bing Xu;66658]</p>\n<p>CXXNET</p>\n<p>[/quote]</p>\n\n<p>I think we would have eyed you suspiciously if you had used anything else, Bing.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66733",
      "postDate": "03/17/2015 16:09:57",
      "content": "<p>Theano</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66768",
      "postDate": "03/17/2015 18:13:01",
      "content": "<p>Started with Torch, hated it. Then looked at CXXNET, which was fine except for the fact that I'm not that fluent in C++ so editing the code was hard. In the end I settled with Lasagne, which I really like because of the fact that it's written in Python + Theano which makes it so easy to modify things. I'm looking forward to seeing the implementation of Sander's team!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66804",
      "postDate": "03/17/2015 19:58:27",
      "content": "<p>Theano / Blocks!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66885",
      "postDate": "03/18/2015 00:45:54",
      "content": "<p>Torch</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66982",
      "postDate": "03/18/2015 15:04:56",
      "content": "<p>CXXNET since that is only thing I tried and used so far.</p>\n<p>I am curious how much CXXNET faster over other python libraries like Theano or Caffe on the same HW running for the same CNN configuration. Has anybody any data on it?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66985",
      "postDate": "03/18/2015 15:21:50",
      "content": "<p>@Ertuka, caffe is not Python, it is C++ library. And it was the fastest for a long time before new fft-based convolution in Theano. When it build with cuDNN, it still one of the fastest.</p>\n<p>https://github.com/soumith/convnet-benchmarks</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "66991",
      "postDate": "03/18/2015 15:51:18",
      "content": "<p>@old-ufo</p>\n<p>Upps my fault. I shouldn't have refer to Caffe since never used it. I&nbsp; installed Theano in windows with partial success and will do in linux soon. Dont' get me wrong, I love Python but currently I have a loptop with GTX 860M and warms up quickly with GPU usage so more inclined thinking that C++ could be more efficient.</p>\n<p>Thanks</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "67009",
      "postDate": "03/18/2015 16:39:15",
      "content": "<p>[quote=old-ufo;66729]</p>\n<p>@Ryan Munion, are you going to share yours solution with Prof. Graham as well?</p>\n<p>[/quote]</p>\n<p>Yes. &nbsp;I will write up something. &nbsp;This was quite the experience for me. &nbsp;I learned a ton about ConvNets.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 66527,
      "author_name": "inversion",
      "author_url": "",
      "post_date": "03/16/2015 21:44:18",
      "content": "<p>NoLearn/Lasagne</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66528,
      "author_name": "dmitrylarko",
      "author_url": "",
      "post_date": "03/16/2015 21:44:56",
      "content": "<p>Lasagne/nolearn</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66531,
      "author_name": "oldufo",
      "author_url": "",
      "post_date": "03/16/2015 21:49:26",
      "content": "<p>Caffe</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66535,
      "author_name": "songgc",
      "author_url": "",
      "post_date": "03/16/2015 21:54:26",
      "content": "<p>Caffe</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66536,
      "author_name": "elenacuoco",
      "author_url": "",
      "post_date": "03/16/2015 21:54:36",
      "content": "<p>Lasagne/nolearn</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66537,
      "author_name": "obaiev",
      "author_url": "",
      "post_date": "03/16/2015 21:59:34",
      "content": "<p>started with Pylearn2, finished with caffe</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66539,
      "author_name": "stevendu",
      "author_url": "",
      "post_date": "03/16/2015 22:01:07",
      "content": "<p>I think he means upvote ... ok anyway Caffe, indeed I don't like it... took half day to install, and one day to find out how to make predictions.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66540,
      "author_name": "ccurro",
      "author_url": "",
      "post_date": "03/16/2015 22:03:26",
      "content": "<p>Torch</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66541,
      "author_name": "inversion",
      "author_url": "",
      "post_date": "03/16/2015 22:06:40",
      "content": "<p>[quote=Steven Du;66539]</p>\n<p>I think he means upvote ... ok anyway Caffe, indeed I don't like it... took half day to install, and one day to find out how to make predictions.</p>\n<p>[/quote]</p>\n<p>LOL . . . yeah.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66543,
      "author_name": "msegala",
      "author_url": "",
      "post_date": "03/16/2015 22:10:08",
      "content": "<p>Lasagne/Nolearn is painless to install and very versatile&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66561,
      "author_name": "lucasbeyer",
      "author_url": "",
      "post_date": "03/16/2015 23:05:07",
      "content": "<p>Plain Theano</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66617,
      "author_name": "hcorey",
      "author_url": "",
      "post_date": "03/17/2015 03:10:56",
      "content": "<p>theano/sklearn</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66619,
      "author_name": "nagadomi",
      "author_url": "",
      "post_date": "03/17/2015 03:18:38",
      "content": "<p>Torch</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66621,
      "author_name": "liangpang",
      "author_url": "",
      "post_date": "03/17/2015 03:20:14",
      "content": "<p>Caffe</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66627,
      "author_name": "telser",
      "author_url": "",
      "post_date": "03/17/2015 03:53:05",
      "content": "<p>Cxxnet + python for image augmentation, looking forward to the update.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66634,
      "author_name": "rightfit",
      "author_url": "",
      "post_date": "03/17/2015 04:58:12",
      "content": "<p>NoLearn/lasagne</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66647,
      "author_name": "oduerr",
      "author_url": "",
      "post_date": "03/17/2015 06:37:54",
      "content": "<p>Caffe...</p>\n<p>but installation and documentation is a pain in the a**. Did not find a way for real time training data augmentation except by brute hacking the cpp-code.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66649,
      "author_name": "only4hj",
      "author_url": "",
      "post_date": "03/17/2015 06:44:53",
      "content": "<p>Lasagne</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66652,
      "author_name": "hochthom",
      "author_url": "",
      "post_date": "03/17/2015 07:09:19",
      "content": "<p>Lasagne / Nolearn</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66658,
      "author_name": "binghsu",
      "author_url": "",
      "post_date": "03/17/2015 07:53:28",
      "content": "<p>CXXNET</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66661,
      "author_name": "selfishgene",
      "author_url": "",
      "post_date": "03/17/2015 08:13:22",
      "content": "<p>mainly Caffe and a little bit of matconvnet</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66671,
      "author_name": "melgor",
      "author_url": "",
      "post_date": "03/17/2015 09:27:29",
      "content": "<p>Caffe, but as I saw the winning solution, I am thinking about Torch or wrapper.&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66672,
      "author_name": "",
      "author_url": "",
      "post_date": "03/17/2015 09:35:16",
      "content": "<p>Caffe and lasagne+nolearn</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66693,
      "author_name": "bigning",
      "author_url": "",
      "post_date": "03/17/2015 13:01:21",
      "content": "<p>cuda_convnet</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66714,
      "author_name": "pikquu",
      "author_url": "",
      "post_date": "03/17/2015 14:50:09",
      "content": "<p>Pylearn2</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66724,
      "author_name": "aharasim",
      "author_url": "",
      "post_date": "03/17/2015 15:45:34",
      "content": "<p>Torch, then CXXNET</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66728,
      "author_name": "iamthep",
      "author_url": "",
      "post_date": "03/17/2015 15:54:05",
      "content": "<p>Spatially Sparse ConvNets&nbsp;</p>\n<p>http://www2.warwick.ac.uk/fac/sci/statistics/staff/academic-research/graham/</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66729,
      "author_name": "oldufo",
      "author_url": "",
      "post_date": "03/17/2015 15:56:56",
      "content": "<p>@Ryan Munion, are you going to share yours solution with Prof. Graham as well?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66730,
      "author_name": "telser",
      "author_url": "",
      "post_date": "03/17/2015 15:57:13",
      "content": "<p>[quote=Bing Xu;66658]</p>\n<p>CXXNET</p>\n<p>[/quote]</p>\n\n<p>I think we would have eyed you suspiciously if you had used anything else, Bing.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66733,
      "author_name": "ahmedmohamedou",
      "author_url": "",
      "post_date": "03/17/2015 16:09:57",
      "content": "<p>Theano</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66768,
      "author_name": "stevenreitsma",
      "author_url": "",
      "post_date": "03/17/2015 18:13:01",
      "content": "<p>Started with Torch, hated it. Then looked at CXXNET, which was fine except for the fact that I'm not that fluent in C++ so editing the code was hard. In the end I settled with Lasagne, which I really like because of the fact that it's written in Python + Theano which makes it so easy to modify things. I'm looking forward to seeing the implementation of Sander's team!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66804,
      "author_name": "lukemetz",
      "author_url": "",
      "post_date": "03/17/2015 19:58:27",
      "content": "<p>Theano / Blocks!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66885,
      "author_name": "pathbinder",
      "author_url": "",
      "post_date": "03/18/2015 00:45:54",
      "content": "<p>Torch</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66982,
      "author_name": "esorar",
      "author_url": "",
      "post_date": "03/18/2015 15:04:56",
      "content": "<p>CXXNET since that is only thing I tried and used so far.</p>\n<p>I am curious how much CXXNET faster over other python libraries like Theano or Caffe on the same HW running for the same CNN configuration. Has anybody any data on it?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66985,
      "author_name": "oldufo",
      "author_url": "",
      "post_date": "03/18/2015 15:21:50",
      "content": "<p>@Ertuka, caffe is not Python, it is C++ library. And it was the fastest for a long time before new fft-based convolution in Theano. When it build with cuDNN, it still one of the fastest.</p>\n<p>https://github.com/soumith/convnet-benchmarks</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 66991,
      "author_name": "esorar",
      "author_url": "",
      "post_date": "03/18/2015 15:51:18",
      "content": "<p>@old-ufo</p>\n<p>Upps my fault. I shouldn't have refer to Caffe since never used it. I&nbsp; installed Theano in windows with partial success and will do in linux soon. Dont' get me wrong, I love Python but currently I have a loptop with GTX 860M and warms up quickly with GPU usage so more inclined thinking that C++ could be more efficient.</p>\n<p>Thanks</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 67009,
      "author_name": "iamthep",
      "author_url": "",
      "post_date": "03/18/2015 16:39:15",
      "content": "<p>[quote=old-ufo;66729]</p>\n<p>@Ryan Munion, are you going to share yours solution with Prof. Graham as well?</p>\n<p>[/quote]</p>\n<p>Yes. &nbsp;I will write up something. &nbsp;This was quite the experience for me. &nbsp;I learned a ton about ConvNets.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "66526": "",
    "66527": "",
    "66528": "",
    "66531": "",
    "66535": "",
    "66536": "",
    "66537": "",
    "66539": "",
    "66540": "",
    "66541": "",
    "66543": "",
    "66561": "",
    "66617": "",
    "66619": "",
    "66621": "",
    "66627": "",
    "66634": "",
    "66647": "",
    "66649": "",
    "66652": "",
    "66658": "",
    "66661": "",
    "66671": "",
    "66672": "",
    "66693": "",
    "66714": "",
    "66724": "",
    "66728": "",
    "66729": "",
    "66730": "",
    "66733": "",
    "66768": "",
    "66804": "",
    "66885": "",
    "66982": "",
    "66985": "",
    "66991": "",
    "67009": ""
  },
  "source": "meta"
}