Spectral Approximate Inference
python & matlab codes used for experiments in "Spectral Approximate Inference" (ICML 2019)
Run evaluate.py with python3
By default, this will compute partition function errors of belief propagation, mean-field approximation, mini-bucket elimination and our spectral approximate inference for pairwise binary models on complete graph of 20 vertices among a range of edge coupling strengths, except for running semi-definite programming of our spectral approximate inference.
To run semi-definite programming, install CVX from http://cvxr.com/cvx/ and run 'compute_sdp_time.m' in matlab_code folder using MATLAB.
This folder contains python classes related with general graphical models and pairwise binary models.
This folder contains python codes for inference algorithms for estimating the partition function (belief propagation, mean-field approximation, mini-bucket elimination and our spectral approximate inference).
This folder contains matlab datasets of pairwise binary graphical models.
This folder contains matlab codes for running semi-definite programming solver used for our spectral approximate inference. Before running 'compute_sdp_time.m', install CVX from http://cvxr.com/cvx/