INSTALL.txt

#########################################
INSTALLATION INSTRUCTIONS FOR DEEPNET

Nitish Srivastava
University of Toronto.
#########################################

(1) DEPENDENCIES
  - Numpy
  - Scipy
  - CUDA 4.2 Toolkit and SDK.
    Install the toolkit and SDK.
    Set an environment variable CUDA_BIN to the path to the /bin directory of
    the cuda installation and CUDA_LIB to the path to the /lib64 (or /lib)
    directory. Also add them to PATH and LD_LIBARAY_PATH.
   
    For example, add the following lines to your ~/.bashrc file
      export CUDA_BIN=/pkgs_local/cuda-4.2/bin
      export CUDA_LIB=/pkgs_local/cuda-4.2/lib64
      export PATH=${CUDA_BIN}:$PATH
      export LD_LIBRARY_PATH=${CUDA_LIB}:$LD_LIBRARY_PATH
      export CPLUS_INCLUDE_PATH=CPLUS_INCLUDE_PATH:/pkgs_local/cuda-4.2/cuda/include

  - Protocol Buffers.
    Available from http://code.google.com/p/protobuf/
    Make sure that the PATH environment variable includes the directory that
    contains the protocol buffer compiler - protoc.
	  For example,
      export PATH=$HOME/local/bin:$PATH

(2) COMPILING CUDAMAT
  DeepNet uses Vlad Mnih's cudamat library. Some additional kernels have been
  added. To compile the library -
  - run make in the cudamat dir.

(3) COMPILING CUDAMAT_CONV
  DeepNet uses Vlad Mnih's cudamat_conv wrapper for Alex Krizhevsky's
  cuda-convnet library. Some additional kernels have been added. To compile the
  library -
  - Edit the paths in cudamat_conv/build.sh.
  - run build.sh in cudamat_conv

(4) SET ENVIRONMENT VARIABLES
  - Add the path to cudamat and cudamat_conv to LD_LIBRARY_PATH. For example if
    DeepNet is located in the home dir,
      export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/deepnet/cudamat:$HOME/deepnet/cudamat_conv
  - Add the path to DeepNet to PYTHONPATH. For example, if DeepNet is located in the
    home dir,
      export PYTHONPATH=$HOME/deepnet:$PYTHONPATH

(5) RUN EXAMPLES
  - Download and extract the MNIST dataset from http://www.cs.toronto.edu/~nitish/deepnet/mnist.tar.gz
    This dataset consists of labelled images of handwritten digits as numpy files.
  - cd to the deepnet/deepnet/examples dir
  - run
    $ python setup_examples.py <path to mnist dataset> <output path>
    This will modify the example models and trainers to use the specified paths.
  - There are examples of different deep learning models. Go to any one and
    execute runall.sh. For example, cd to deepnet/deepnet/examples/rbm and execute
    $ ./runall.sh
    This should start training an RBM model.
  
DeepNet has been tested on Ubuntu 12.04 using CUDA 4.2 on a variety of NVIDIA
GPUs (GTX-280, GTX-580, GTX-690, M2090, K-20x).

If you have trouble installing/running DeepNet, please create a relevant post on GitHub.