Skip to main content

Optimized (and optionally gpu enhaced) fitting of Gaussian Mixture Models

Project description

dpmix is a library for understanding posterior distributions for Dirichlet and heirarchical Dirichlet mixtures of normal distributions represented by truncated stick breaking.


  • NumPy

  • SciPy

  • Cython

  • PyCUDA

  • cyarma

  • cyrand

  • scikits.cuda

  • gpustats

  • mpi4py

Installation and testing

Install via

python install

To test, run the scripts in the “test” subfolder.


Check out the class docstrings for more info.


The multigpu facilities are developed using MPI. Therefore, using multiple machines is possible. However, note that the machines must be configured the same way. (Python)

Running the code on multiple machines requires mpiexec:

mpiexec -hostfile my_hosts -np 1 python tests/ --gpu MPI

Where the my_hosts file looks like

host1 slots=3
host2 slots=2

I’m assuming here that the master instance of python is running on host1 and that host1 and host2 have 2 GPUs each. Note, an extra slot needs to be reserved for the master on host1. Furthermore, we need to specify which devices to use on each host. The gpu argument in the class constructors must be a dictionary like

gpu={'host1': [0,1], 'host2': [0,1]}

The keys must match the result of a call to os.uname() to get the host string.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

dpmix-0.3.tar.gz (130.1 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page