This is a test run of the QKM algorithm (see Still, et. al. "Geometric Clustering using the Information Bottleneck method" http://books.nips.cc/papers/files/nips16/NIPS2003_LT13.pdf).
Data point cluster membership probabilities are encoded in color, making this fun to watch.
Parameters:
4 clusters, 2500 points; random initial centers are blue, solution centers are red.
initial temperature=0.5
iterations (converges at around 18)=100
initial variance=200
Link to this comment:
All Comments (0)