Bayesian Nonparametric Small-Variance Asymptotic Clustering

This is a library of Bayesian nonparametric small-variance asymptotic clustering algorithms: DP-means, Dynamic means, DP-vMF-means, DDP-vMF-means. For comparison reasons the library also implements k-means as well as spherical k-means.

This library comes with an executable that allows batch clustering using DP-vMF-means, DP-means, spherical k-means and k-means.

The pure python implementation of DP-vMF-means shows the simplicity of the algorithm.

For an example of using DDP-vMF-means refer to rtDDPvMF, which relies on this package's dpMMlowVar library to perform real-time directional segmentation from Kinect RGB-D streams using DDP-vMF-means.

If you use DP-vMF-means or DDP-vMF-means please cite:

Julian Straub, Trevor Campbell, Jonathan P. How, John W. Fisher III. 
"Small-Variance Nonparametric Clustering on the Hypersphere", In CVPR,
2015.

If you use Dynamic-means please cite:

T. Campbell, M. Liu, B. Kulis, J. How, and L. Carin. "Dynamic
Clustering via Asymptotics of the Dependent Dirichlet Process Mixture".
In Advances in Neural Information Processing Systems (NIPS), 2013.

Dependencies

This code is dependent on Eigen3, Boost, CUDA, OpenCV, OpenMP and PCL. It has been tested on Ubuntu 14.04 with

• Eigen3 (3.0.5)
• Boost (1.54)
• CUDA (6.5)
• OpenCV (2.4)
• PCL (1.7)

Install

This package uses the pods build system. Used widely at CSAIL MIT the build system makes it easy to break up software projects into small packages that can be checked out and compiled automatically (see below).

• Linux:

  Install Eigen3, Boost, OpenCV, and PCL

  sudo apt-get install libeigen3-dev libboost-dev libopencv-dev libpcl-1.7-all-dev
  

  Install the appropriate CUDA version matching with your nvidia drivers. On our machines we use nvidia-340-dev with libcuda1-340 cuda-6-5 cuda-toolkit-6-5

  Clone this repository and compile the code:

  git clone git@github.com:jstraub/dpMMlowVar; cd dpMMlowVar;
  make checkout; make configure; make -j6; make install;
  

  Note that this will checkout several other necessary repositories. To update all repositories run

  make update; make configure; make -j6; make install;
  

Getting Started

After you have compiled the code you can run clustering of the surface normals of an example data set in the ./data/ folder by running:

cd ./python; python dpView.py

Note that the extraction of surface normals necessitates matlab. An alternative is to directly run segmentation from Kinect RGB-D frames using the rtDDPvMF package.

The pure python implementation of DP-vMF-means and of DP-means shows the simplicity of the nonparametric algorithms.

Usage

./dpMMlowVarCluster -h
Allowed options:
  -h [ --help ]         produce help message
  --seed arg            seed for random number generator
  -N [ --N ] arg        number of input datapoints
  -D [ --D ] arg        number of dimensions of the data
  -T [ --T ] arg        iterations
  -a [ --alpha ] arg    alpha parameter of the DP (if single value assumes all 
                        alpha_i are the same
  -K [ --K ] arg        number of initial clusters 
  --base arg            which base measure to use (only spkm, kmeans, 
                        DPvMFmeans right now)
  -p [ --params ] arg   parameters of the base measure
  -i [ --input ] arg    path to input dataset .csv file (rows: dimensions; 
                        cols: different datapoints)
  -o [ --output ] arg   path to output labels .csv file (rows: time; cols: 
                        different datapoints)
  --mlInds              output ml indices
  --centroids           output centroids of clusters
  --silhouette          output average silhouette
  --shuffle             shuffle the data before processing

Examples:

• DPvMFmeans clustering: First compute the lambda parameter of DPvMFmeans as lambda = cos(angleInRadians) - 1. so for example lambda = -0.06 for an angle of 20deg.

./dpMMlowVarCluster -N 10000 -D 3 --base DPvMFmeans -p -0.06 -i ./data/rndSphereDataIwUncertain.csv -o test

Contributors

Julian Straub and Trevor D. Campbell