I am wondering whether anyone has been able to work on this competition on a Window OS? I decided early on to use Python instead of R which I am more used to. However, I could not install key components like Theano or Tensorflow in Windows. So I loaded Docker and set up a VM onto which I installed python and keras with a tensorflow backend. Every step has been a struggle since I am not too familiar with linux and VMs. Now I am challenged to get an editor running so I can work with scripts and I am concerned compromises I had to make along the way (like CPU instead of GPU) will make training too time consuming.
\n\nHas anyone successfully configured a windows box for this competition? If so, is there a simpler approach than the one I am taking? Is there a package(s) that can support CNNs in Windows?
", "messages": [ { "id": "118998", "postDate": "05/06/2016 16:00:46", "content": "I am wondering whether anyone has been able to work on this competition on a Window OS? I decided early on to use Python instead of R which I am more used to. However, I could not install key components like Theano or Tensorflow in Windows. So I loaded Docker and set up a VM onto which I installed python and keras with a tensorflow backend. Every step has been a struggle since I am not too familiar with linux and VMs. Now I am challenged to get an editor running so I can work with scripts and I am concerned compromises I had to make along the way (like CPU instead of GPU) will make training too time consuming.
\n\nHas anyone successfully configured a windows box for this competition? If so, is there a simpler approach than the one I am taking? Is there a package(s) that can support CNNs in Windows?
", "rawMarkdown": "I am wondering whether anyone has been able to work on this competition on a Window OS? I decided early on to use Python instead of R which I am more used to. However, I could not install key components like Theano or Tensorflow in Windows. So I loaded Docker and set up a VM onto which I installed python and keras with a tensorflow backend. Every step has been a struggle since I am not too familiar with linux and VMs. Now I am challenged to get an editor running so I can work with scripts and I am concerned compromises I had to make along the way (like CPU instead of GPU) will make training too time consuming.\r\n\r\nHas anyone successfully configured a windows box for this competition? If so, is there a simpler approach than the one I am taking? Is there a package(s) that can support CNNs in Windows?", "votes": null }, { "id": "119005", "postDate": "05/06/2016 16:30:40", "content": "I got something working on Windows yesterday. If you are willing to work on Keras/Theano, here is something that worked for me: https://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/
\n\nAs far as I know, there is no Tensorflow support on Windows yet.
\n\nHope this helps.
", "rawMarkdown": "I got something working on Windows yesterday. If you are willing to work on Keras/Theano, here is something that worked for me: https://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/\r\n\r\nAs far as I know, there is no Tensorflow support on Windows yet.\r\n\r\nHope this helps.", "votes": null }, { "id": "119023", "postDate": "05/06/2016 18:42:52", "content": "Thanks, I'll try that post again. When I first tried it, it might have failed because python 2.7 was my secondary and not primary environment and I don't have a Nvidia GPU, but now that I understand better what's going on, maybe it will work.
", "rawMarkdown": "Thanks, I'll try that post again. When I first tried it, it might have failed because python 2.7 was my secondary and not primary environment and I don't have a Nvidia GPU, but now that I understand better what's going on, maybe it will work.", "votes": null }, { "id": "119070", "postDate": "05/07/2016 00:41:28", "content": "Maybe,you can use Mxnet
", "rawMarkdown": "Maybe,you can use Mxnet", "votes": null }, { "id": "119080", "postDate": "05/07/2016 02:16:59", "content": "There is new Microsoft's CNTK. Speed benchmark is quite impressive, faster than Theano & TensorFlow. But afaik there is no Python binding yet so far.
\n\nAnd, Caffe's Windows branch. Page: http://caffe.berkeleyvision.org/ .\nThis one has Python binding (pycaffe).
", "rawMarkdown": "There is new [Microsoft's CNTK][1]. Speed benchmark is quite impressive, faster than Theano & TensorFlow. But afaik there is no Python binding yet so far. \r\n\r\nAnd, Caffe's [Windows branch][2]. Page: http://caffe.berkeleyvision.org/ .\r\nThis one has Python binding (pycaffe). \r\n\r\n [1]: https://github.com/Microsoft/CNTK\r\n [2]: https://github.com/BVLC/caffe/tree/windows", "votes": null }, { "id": "119418", "postDate": "05/10/2016 05:14:00", "content": "[quote=kboom;118998]
\n\nHas anyone successfully configured a windows box for this competition? If so, is there a simpler approach than the one I am taking? Is there a package(s) that can support CNNs in Windows?
\n\n[/quote]
\n\nI am able to use Win7. It is a bit of headache, but after I am past initial hurdles of installing Theano and Keras (which, honestly, weren't too bad with anaconda) it doesn't give me any problems. It is worth investing some time to installing Theano and Keras, imho, as VM would create too much overhead. Working without GPU would also be quite slow.
", "rawMarkdown": "[quote=kboom;118998]\r\n\r\nHas anyone successfully configured a windows box for this competition? If so, is there a simpler approach than the one I am taking? Is there a package(s) that can support CNNs in Windows?\r\n\r\n[/quote]\r\n\r\nI am able to use Win7. It is a bit of headache, but after I am past initial hurdles of installing Theano and Keras (which, honestly, weren't too bad with anaconda) it doesn't give me any problems. It is worth investing some time to installing Theano and Keras, imho, as VM would create too much overhead. Working without GPU would also be quite slow.", "votes": null }, { "id": "119431", "postDate": "05/10/2016 07:44:54", "content": "I'm using Theano (GPU) with Windows 7. I was able to install Keras on the top of it as well. Actually I installed it some time ago and don't remember everything except it wasn't easy )
\n\nI started from WinPython (version 3.4):
\n\n\n\nAnd install/update some modules which can be easily installed from WHL-files with pip in Windows: http://www.lfd.uci.edu/~gohlke/pythonlibs/
\n\nAs I remeber for Theano you need to:
\n\n1) Install CuDNN
\n\n2) Install some C++ compiler (http://tdm-gcc.tdragon.net/) and have Visual Studio (not sure if needed)
\n\n3) Create config file in path like this:
\n\nC:\\Users\\<User_name_here>\\.theanorc\nContents of file:
\n\n[global]\nfloatX = float32\ndevice = gpu0\ncxx = C:\\Programs\\TDM-GCC-64\\bin\\g++.exe\nforce_device = True\n\n[cuda]\nroot=C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v7.5\n\n[lib]\ncnmem = 0.85\n\n[nvcc]\nfastmath = True\ncompiler_bindir=C:\\Program Files (x86)\\Microsoft Visual Studio 12.0\\VC\\bin\\amd64\nSome notes:
\n\nYou need Visual Studios 12.0 and include it, plus the C++ compiler in your computer path environment.
", "rawMarkdown": "You need Visual Studios 12.0 and include it, plus the C++ compiler in your computer path environment.", "votes": null } ], "comments": [ { "id": 119005, "author_name": "prith189", "author_url": "", "post_date": "05/06/2016 16:30:40", "content": "I got something working on Windows yesterday. If you are willing to work on Keras/Theano, here is something that worked for me: https://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/
\n\nAs far as I know, there is no Tensorflow support on Windows yet.
\n\nHope this helps.
", "votes": null, "replies": [] }, { "id": 119023, "author_name": "kboomer", "author_url": "", "post_date": "05/06/2016 18:42:52", "content": "Thanks, I'll try that post again. When I first tried it, it might have failed because python 2.7 was my secondary and not primary environment and I don't have a Nvidia GPU, but now that I understand better what's going on, maybe it will work.
", "votes": null, "replies": [] }, { "id": 119070, "author_name": "juguangliang", "author_url": "", "post_date": "05/07/2016 00:41:28", "content": "Maybe,you can use Mxnet
", "votes": null, "replies": [] }, { "id": 119080, "author_name": "aldente", "author_url": "", "post_date": "05/07/2016 02:16:59", "content": "There is new Microsoft's CNTK. Speed benchmark is quite impressive, faster than Theano & TensorFlow. But afaik there is no Python binding yet so far.
\n\nAnd, Caffe's Windows branch. Page: http://caffe.berkeleyvision.org/ .\nThis one has Python binding (pycaffe).
", "votes": null, "replies": [] }, { "id": 119418, "author_name": "alexlzzz", "author_url": "", "post_date": "05/10/2016 05:14:00", "content": "[quote=kboom;118998]
\n\nHas anyone successfully configured a windows box for this competition? If so, is there a simpler approach than the one I am taking? Is there a package(s) that can support CNNs in Windows?
\n\n[/quote]
\n\nI am able to use Win7. It is a bit of headache, but after I am past initial hurdles of installing Theano and Keras (which, honestly, weren't too bad with anaconda) it doesn't give me any problems. It is worth investing some time to installing Theano and Keras, imho, as VM would create too much overhead. Working without GPU would also be quite slow.
", "votes": null, "replies": [] }, { "id": 119431, "author_name": "zfturbo", "author_url": "", "post_date": "05/10/2016 07:44:54", "content": "I'm using Theano (GPU) with Windows 7. I was able to install Keras on the top of it as well. Actually I installed it some time ago and don't remember everything except it wasn't easy )
\n\nI started from WinPython (version 3.4):
\n\n\n\nAnd install/update some modules which can be easily installed from WHL-files with pip in Windows: http://www.lfd.uci.edu/~gohlke/pythonlibs/
\n\nAs I remeber for Theano you need to:
\n\n1) Install CuDNN
\n\n2) Install some C++ compiler (http://tdm-gcc.tdragon.net/) and have Visual Studio (not sure if needed)
\n\n3) Create config file in path like this:
\n\nC:\\Users\\<User_name_here>\\.theanorc\nContents of file:
\n\n[global]\nfloatX = float32\ndevice = gpu0\ncxx = C:\\Programs\\TDM-GCC-64\\bin\\g++.exe\nforce_device = True\n\n[cuda]\nroot=C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v7.5\n\n[lib]\ncnmem = 0.85\n\n[nvcc]\nfastmath = True\ncompiler_bindir=C:\\Program Files (x86)\\Microsoft Visual Studio 12.0\\VC\\bin\\amd64\nSome notes:
\n\nYou need Visual Studios 12.0 and include it, plus the C++ compiler in your computer path environment.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "118998": "I am wondering whether anyone has been able to work on this competition on a Window OS? I decided early on to use Python instead of R which I am more used to. However, I could not install key components like Theano or Tensorflow in Windows. So I loaded Docker and set up a VM onto which I installed python and keras with a tensorflow backend. Every step has been a struggle since I am not too familiar with linux and VMs. Now I am challenged to get an editor running so I can work with scripts and I am concerned compromises I had to make along the way (like CPU instead of GPU) will make training too time consuming.\r\n\r\nHas anyone successfully configured a windows box for this competition? If so, is there a simpler approach than the one I am taking? Is there a package(s) that can support CNNs in Windows?", "119005": "I got something working on Windows yesterday. If you are willing to work on Keras/Theano, here is something that worked for me: https://datanoord.com/2016/02/01/setup-a-deep-learning-environment-on-windows-theano-keras-with-gpu-enabled/\r\n\r\nAs far as I know, there is no Tensorflow support on Windows yet.\r\n\r\nHope this helps.", "119023": "Thanks, I'll try that post again. When I first tried it, it might have failed because python 2.7 was my secondary and not primary environment and I don't have a Nvidia GPU, but now that I understand better what's going on, maybe it will work.", "119070": "Maybe,you can use Mxnet", "119080": "There is new [Microsoft's CNTK][1]. Speed benchmark is quite impressive, faster than Theano & TensorFlow. But afaik there is no Python binding yet so far. \r\n\r\nAnd, Caffe's [Windows branch][2]. Page: http://caffe.berkeleyvision.org/ .\r\nThis one has Python binding (pycaffe). \r\n\r\n [1]: https://github.com/Microsoft/CNTK\r\n [2]: https://github.com/BVLC/caffe/tree/windows", "119418": "[quote=kboom;118998]\r\n\r\nHas anyone successfully configured a windows box for this competition? If so, is there a simpler approach than the one I am taking? Is there a package(s) that can support CNNs in Windows?\r\n\r\n[/quote]\r\n\r\nI am able to use Win7. It is a bit of headache, but after I am past initial hurdles of installing Theano and Keras (which, honestly, weren't too bad with anaconda) it doesn't give me any problems. It is worth investing some time to installing Theano and Keras, imho, as VM would create too much overhead. Working without GPU would also be quite slow.", "119431": "I'm using Theano (GPU) with Windows 7. I was able to install Keras on the top of it as well. Actually I installed it some time ago and don't remember everything except it wasn't easy )\r\n\r\nI started from WinPython (version 3.4): \r\n\r\nhttps://winpython.github.io/\r\n\r\nAnd install/update some modules which can be easily installed from WHL-files with pip in Windows: http://www.lfd.uci.edu/~gohlke/pythonlibs/\r\n\r\nAs I remeber for Theano you need to:\r\n\r\n1) Install CuDNN\r\n\r\n2) Install some C++ compiler (http://tdm-gcc.tdragon.net/) and have Visual Studio (not sure if needed)\r\n\r\n3) Create config file in path like this:\r\n\r\n C:\\Users\\