dimreduce4gpu

Dimensionality reduction (dimreduce) on GPUs (4gpu)

• Current Algorithms:
  • PCA
  • Truncated SVD

Building

Build Environment

• NOTE: dimreduce4gpu is tested up to CUDA 9.0

• Linux machine w/ GCC4.9+ and CMake installed.

• Python 3.6.

For virtualenv and ubuntu 16.04:

apt-get -y --no-install-recommends  install \
    python3.6 \
    python3.6-dev \
    virtualenv \
    python3-pip
virtualenv --python=python3.6 .venv
pip install setuptools --no-cache-dir
. .venv/bin/activate

• Add to .bashrc or your own environment (e.g.):

export CUDA_HOME=/usr/local/cuda
export PATH=$CUDA_HOME/bin:$PATH
export LD_LIBRARY_PATH_MORE=/home/$USER/lib/:$CUDA_HOME/lib64/:$CUDA_HOME/lib/:$CUDA_HOME/lib64:$CUDA_HOME/extras/CUPTI/lib64
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$LD_LIBRARY_PATH_MORE
export CUDADIR=/usr/local/cuda/include/

• To compile everything, install the Python interface, and run a simple test:

git clone --recursive git@github.com:navdeep-G/dimreduce4gpu.git
cd dimreduce4gpu
virtualenv -p python3.6 env
make

dimreduce4gpu usage in other projects

• This library (or a variant of it) is used in the following open source projects:
  • H2O4GPU(A collection of GPU solvers by H2O.ai )
    • Usage:
      • Truncated SVD
      • Principal Components Analysis