MITHaystack / cuda_mcmc

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

cuda_mcmc

The cuda_mcmc software is designed to greatly accelerate solving various optimization problems using the NVIDIA graphics processing units (GPU) with the CUDA computing platform. We use a version of the Markov chain Monte-Carlo (MCMC) algorithm based on the Metropolis-Hastings sampler with replica exchange (MCMC-RE). The detailed description of the algorithm with its mapping on the CUDA GPU architecture including the user instructions to its installation and writing custom optimization/equation solving applications is provided in the doc/ directory. Several examples of the code usage are given in the examples/ directory.

To start using CUDA MCMC software, first enter its containing directory (usually ~/cuda_mcmc) and issue the command

$ make

It will create the directories ~/lib64/python, and ~/bin, and setup the environment for them by adding to the end of the ~/.bashrc file the lines

export PATH=$PATH:~/bin
export PYTHONPATH=$PYTHONPATH:~/lib64/python

After make finishes, update the environment variables PATH and PYTHONPATH issuing the command

$ source ~/.bashrc

Or simply open a new console window. This will automatically run ~/.bashrc script and set the environment variables.

About


Languages

Language:Python 67.4%Language:Cuda 24.8%Language:C 6.4%Language:Makefile 1.4%