{ "id": 4528, "title": "Corruptor level in Autoencoder", "url": "/competitions/challenges-in-representation-learning-the-black-box-learning-challenge/discussion/4528", "author_name": "", "post_date": "2013-05-07T21:57:08.337Z", "votes": null, "comment_count": 2, "views": 1745, "content": "

I'm experimenting with several level of corruption_level (stdev param in gaussian corruptor) and the model trace is exactly the same.

\r\n

This is the block I'm using:

\r\n
    "model": !obj:pylearn2.models.autoencoder.HigherOrderContractiveAutoencoder {
        "nvis" : 1875,
        "nhid" : 200,
        "irange" : 0.05,
        "corruptor": !obj:pylearn2.corruption.GaussianCorruptor {
            "stdev": 0.1,
        },
        "num_corruptions": 1,
        "act_enc": 'sigmoid',
        "act_dec": 'sigmoid',
    },
\r\n
Someone have obtained differences changing stdev in GaussianCorruptor?
", "messages": [ { "id": "23999", "postDate": "05/07/2013 21:57:08", "content": "

I'm experimenting with several level of corruption_level (stdev param in gaussian corruptor) and the model trace is exactly the same.

\r\n

This is the block I'm using:

\r\n
    "model": !obj:pylearn2.models.autoencoder.HigherOrderContractiveAutoencoder {
        "nvis" : 1875,
        "nhid" : 200,
        "irange" : 0.05,
        "corruptor": !obj:pylearn2.corruption.GaussianCorruptor {
            "stdev": 0.1,
        },
        "num_corruptions": 1,
        "act_enc": 'sigmoid',
        "act_dec": 'sigmoid',
    },
\r\n
Someone have obtained differences changing stdev in GaussianCorruptor?
", "rawMarkdown": "", "votes": null }, { "id": "24003", "postDate": "05/08/2013 02:56:00", "content": "Can you e-mail me the full yaml file?", "rawMarkdown": "", "votes": null }, { "id": "24026", "postDate": "05/08/2013 17:30:52", "content": "

[quote=Ian Goodfellow;24003]

\r\n

Can you e-mail me the full yaml file?

\r\n

[/quote]

\r\n

The problem is gpu related (theano.sandbox.cuda). I can't get my gpu run under W7 64 with theano. I'm trying to fix it.

\r\n

EDIT.

\r\n

The problem is CL.exe. I haven't VisualStudio (and I don't want it). CL is a pain if you haven't the microsoft suit installed (I've intalled only C++ without the VS IDE and this is the problem).

\r\n

I've seen theano installed minGW. Now the question is why CL is neccesary?

\r\n

Could I configure the compiler with minGW?

\r\n

Ian has expressed he haven't windows experience, so someone has configured the setup Theano - Pylearn2 - CUDA under W7 64 bits and could share the trick?

\r\n

EDIT

\r\n

Found here:

\r\n

http://stackoverflow.com/questions/2970493/cuda-linking-error-visual-express-2008-nvcc-fatal-due-to-null-configuratio

\r\n

I solved the problem by 1. installing Windows SDK (don't forget to choose all x64 options for 64 bit OS)\r\n

\r\n

2. include "c:\\Program Files (x86)\\Microsoft Visual Studio 10.0\\VC\\bin\\amd64" in PATH\r\n

\r\n

3. create file vcvars64.bat inside directory "c:\\Program Files (x86)\\Microsoft Visual Studio 10.0\\VC\\bin\\amd64" with following content: call "C:\\Program Files\\Microsoft SDKs\\Windows\\v7.1\\Bin\\SetEnv.cmd" /x64

\r\n

 

\r\n

Really I'm going to have to install Windows 7 SDK for compiling CUDA in Theano? This is a nightmare.

\r\n

EDIT

\r\n

SOLVED! :This weekend I'll install a partition with Ubuntu...

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 24003, "author_name": "iangoodfellow", "author_url": "", "post_date": "05/08/2013 02:56:00", "content": "Can you e-mail me the full yaml file?", "votes": null, "replies": [] }, { "id": 24026, "author_name": "blindape", "author_url": "", "post_date": "05/08/2013 17:30:52", "content": "

[quote=Ian Goodfellow;24003]

\r\n

Can you e-mail me the full yaml file?

\r\n

[/quote]

\r\n

The problem is gpu related (theano.sandbox.cuda). I can't get my gpu run under W7 64 with theano. I'm trying to fix it.

\r\n

EDIT.

\r\n

The problem is CL.exe. I haven't VisualStudio (and I don't want it). CL is a pain if you haven't the microsoft suit installed (I've intalled only C++ without the VS IDE and this is the problem).

\r\n

I've seen theano installed minGW. Now the question is why CL is neccesary?

\r\n

Could I configure the compiler with minGW?

\r\n

Ian has expressed he haven't windows experience, so someone has configured the setup Theano - Pylearn2 - CUDA under W7 64 bits and could share the trick?

\r\n

EDIT

\r\n

Found here:

\r\n

http://stackoverflow.com/questions/2970493/cuda-linking-error-visual-express-2008-nvcc-fatal-due-to-null-configuratio

\r\n

I solved the problem by 1. installing Windows SDK (don't forget to choose all x64 options for 64 bit OS)\r\n

\r\n

2. include "c:\\Program Files (x86)\\Microsoft Visual Studio 10.0\\VC\\bin\\amd64" in PATH\r\n

\r\n

3. create file vcvars64.bat inside directory "c:\\Program Files (x86)\\Microsoft Visual Studio 10.0\\VC\\bin\\amd64" with following content: call "C:\\Program Files\\Microsoft SDKs\\Windows\\v7.1\\Bin\\SetEnv.cmd" /x64

\r\n

 

\r\n

Really I'm going to have to install Windows 7 SDK for compiling CUDA in Theano? This is a nightmare.

\r\n

EDIT

\r\n

SOLVED! :This weekend I'll install a partition with Ubuntu...

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "23999": "", "24003": "", "24026": "" }, "source": "meta" }