The MATLAB code skeleton used in the paper "Support Consistency of Direct Sparse-Change Learning in Markov Networks".
There are only two steps to produce the "probability of success" plot used in the paper.
- demo_POS
It runs the "KLIEP algorithm" and learns the changes between two (lattice) MNs with the difference of 4 edges.
The script uses "parfor" to run 300 times using different random samples in parallel. - plot_POS:
With the results generated from the first step, it now plots the "probability of success" curves shown in the paper.
Reference:
Liu, S., Suzuki, T., Relator R., Sese J., Sugiyama, M., Fukumizu, K.,
Support Consistency of Direct Sparse-Change Learning in Markov Networks
Presented at NIPS workshop on Transfer and Multi-task learning: Theory Meets Practice
Proceedings of Twenty-Ninth AAAI Conference on Artificial Intelligence (AAAI2015)
, pp.2785-2791, 2015.
To appear in Annals of Statistics, 2016, arxiv.
Contact: liu@ism.ac.jp