This software package contains a Barnes-Hut implementation of the t-SNE algorithm. The implementation is described in this paper.

Note: This repository is a version of t-SNE modified to support ongoing research. It may be slightly slower than the original. If you're just trying to run t-SNE, check the original repository that we forked from.

Installation

On Linux or OS X, compile the source using the following command:

g++ sptree.cpp tsne.cpp tsne_main.cpp -o bh_tsne -O2

The executable will be called bh_tsne.

On Windows using Visual C++, do the following in your command line:

• Find the vcvars64.bat file in your Visual C++ installation directory. This file may be named vcvars64.bat or something similar. For example:

  // Visual Studio 12
  "C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin\amd64\vcvars64.bat"

  // Visual Studio 2013 Express:
  C:\VisualStudioExp2013\VC\bin\x86_amd64\vcvarsx86_amd64.bat

• From cmd.exe, go to the directory containing that .bat file and run it.

• Go to bhtsne directory and run:

  nmake -f Makefile.win all

The executable will be called windows\bh_tsne.exe.

Usage

The code comes with wrappers for Matlab and Python. These wrappers write your data to a file called data.dat, run the bh_tsne binary, and read the result file result.dat that the binary produces. There are also external wrappers available for Torch, R, and Julia. Writing your own wrapper should be straightforward; please refer to one of the existing wrappers for the format of the data and result files.

Demonstration of usage in Matlab:

filename = websave('mnist_train.mat', 'https://github.com/awni/cs224n-pa4/blob/master/Simple_tSNE/mnist_train.mat?raw=true');
load(filename);
numDims = 2; pcaDims = 50; perplexity = 50; theta = .5; alg = 'svd';
map = fast_tsne(digits', numDims, pcaDims, perplexity, theta, alg);
gscatter(map(:,1), map(:,2), labels');

Demonstration of usage in Python:

import numpy as np
import bhtsne

data = np.loadtxt("mnist2500_X.txt", skiprows=1)

embedding_array = bhtsne.run_bh_tsne(data, initial_dims=data.shape[1])

Python Wrapper

Usage:

python bhtsne.py [-h] [-d NO_DIMS] [-p PERPLEXITY] [-t THETA]
                  [-r RANDSEED] [-n INITIAL_DIMS] [-v] [-i INPUT]
                  [-o OUTPUT] [--use_pca] [--no_pca] [-m MAX_ITER]

Below are the various options the wrapper program bhtsne.py expects:

• -h, --help show this help message and exit
• -d NO_DIMS, --no_dims NO_DIMS
• -p PERPLEXITY, --perplexity PERPLEXITY
• -t THETA, --theta THETA
• -r RANDSEED, --randseed RANDSEED
• -n INITIAL_DIMS, --initial_dims INITIAL_DIMS
• -v, --verbose
• -i INPUT, --input INPUT: the input file, expects a TSV with the first row as the header.
• -o OUTPUT, --output OUTPUT: A TSV file having each row as the d dimensional embedding.
• --use_pca
• --no_pca
• -m MAX_ITER, --max_iter MAX_ITER