ucb-ischool / mvkdr

Multi-view Spectral Clustering on Conflicting Views

MvKDR

This repository contains Python implementations of the algorithm MvKDR described in "Multi-view Spectral Clustering on Conflicting Views", which is appearing in ECML/PKDD 2017, SKOPJE, MACEDONIA

Dependencies

Python 2.7, modern versions of numpy, scipy, pandas, scikit-learn. All of them available via pip.

Usage

The implementation of MvKDR is in code/mvkdr.py (currently only two views are supported)

Km_label, km_obj = mvkdr(X1, X2, sigma1, sigma2, lambda1, lambda2, seed)

input:

X1: a n X p1 numpy matrix of n samples and p1 feaures in view 1

X2: a n X p2 numpy matrix of n samples and p2 features in view 2

sigma1:  a float for sigma for gaussian kernel for X1, should be set to the median of pairwise distance of X1

sigma2:  a float for sigma for gaussian kernel for X2, should be set to the median of pairwise distance of X2

lambda1:  a float indictes the regularization parameter of agreement between subspace projection 

lambda2:  a float indictes the regularization parameter of disagreement between alternative subspace projection

seed:   an integer indicates the seed for initialization

Output:

km_label:  a vector of size n for clustering label produced by k-means

km_obj: a float of k-means objective value

Contact

Any questions can be directed to:

• Xiao He: xiao.he [at] bsse.ethz.ch