t-viSNE: Interactive Assessment and Interpretation of t-SNE Projections

This Git repository contains the code that accompanies a research publication so-called "t-viSNE: Interactive Assessment and Interpretation of t-SNE Projections". The details of the experiments and the research outcome are described in this paper.

Note: This repository contains a frozen version (commit id: 127) that matches the paper's implementation. However, we plan to improve the implementation in the future.

Data Sets

The data sets are available online from the UCI Machine Learning Repository. We use the Iris data set, Breast Cancer Wisconsin (Original) data set, Pima Indians Diabetes data set, and the SPECTF Heart data set. All data sets are transformed in comma separated format (csv).

Requirements

For the backend:

• Python3
• Flask

For the frontend:

• D3.js
• Plotly.js

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 Python:

# first terminal: hosting the visualization side (client)
python3 -m http.server # for Python3
#or 
python -m SimpleHTTPServer 8000 # for Python2

# second terminal: hosting the computational side (server)
FLASK_APP=tsneGrid.py flask run