Compares ADMM and an interior-point method on MAP inference for HL-MRFs.
Groovy Python Shell
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
data
src/main
README.md
pom.xml
run.sh

README.md

admm-speed-test

Code for comparing ADMM and an interior-point method, as presented in Hinge-Loss Markov Random Fields and Probabilistic Soft Logic and my Ph.D. dissertation. It also contains the data and setup for the paper Scaling MPE Inference for Constrained Continuous Markov Random Fields with Consensus Optimization, but this code tests the improved version of the inference algorithm for piecewise-quadratic problems.

Please cite this work as

@article{bach:arxiv15,
 Title = {Hinge-Loss Markov Random Fields and Probabilistic Soft Logic},
 Author = {Bach, Stephen H. and Broecheler, Matthias and Huang, Bert and Getoor, Lise},
 Volume = {arXiv:1505.04406 [cs.LG]},
 Year = {2015}}

Instructions

Prerequisites

This software depends on Java 6+ and Maven 3. Python (>=2.7) is also required to process the results.

Running the experiment also requires installing the MOSEK add-on for PSL. Please follow the instructions.

PSL Library

The algorithms for this experiment are implemented in the PSL library, version 1.2. It will be downloaded by Maven automatically.

Executing

The experiment can be run by executing run.sh.