uafshahid / tsne

A python wrapper for Barnes-Hut tsne

Python-TSNE

Python library containing T-SNE algorithms, with significant performance improvement over the original projects from which this fork was created.

Note: Scikit-learn v0.17 includes TSNE algorithms and you should probably be using them instead of this.

Algorithms

Barnes-Hut-SNE

A python (cython) wrapper for Barnes-Hut-SNE aka fast-tsne.

I basically took osdf's code and made it pip compliant.

Requirements

• numpy > =1.7.1
• scipy >= 0.12.0
• cython >= 0.19.1
• cblas or openblas. Tested version is v0.2.5 and v0.2.6 (not necessary for OSX).
• OpenMp

Installation

You can install the package from the Github repository:

pip install git+https://github.com/rappdw/tsne.git

Or using docker (to build a wheel for linux):

$ bin/build-image jenkins
$ bin/run-image jenkins
$ (in container) python setup.py bdist_wheel

# This will generate a wheel in your project's dist directory

Or on OSX

First install gcc (for openmp support) via: brew install gcc --without-multilib. Then install CMake. Once that's in place you can run pip or setup.py to build a wheel or install locally, e.g.

• export PATH="/Applications/CMake.app/Contents/bin/:$PATH"; export CC=/usr/local/bin/gcc-6; export CXX=/usr/local/bin/g++-6; python setup.py bdist_wheel
• export PATH="/Applications/CMake.app/Contents/bin/:$PATH"; export CXX=/usr/local/bin/g++-6; export CC=/usr/local/bin/gcc-6; pip install -e .

Usage

Basic usage:

from tsne import bh_sne
X_2d = bh_sne(X)

Or, the wheels also contain an executable that can be used from the command-line as described in the original project.

Examples

• Iris
• MNIST
• word2vec on presidential speeches via @prateekpg2455

More Information

See Barnes-Hut-SNE (2013), L.J.P. van der Maaten. It is available on arxiv.

Mulit-core

see: Python 3 - Multicore

Performance

This fork of the orignal project has a number of performance improvements resulting in an order of magnitude performance improvement when running on multi-core systems. See tsne-pref-test for performance comparisions of various branches of BH-t-SNE.