Alpha Belief Propagation

Code for our paper: $\alpha$ Belief Propagation for Approximate Inference

Environment

Create a virtual environment with a python2 interpreter at 'path/to/your/evn/'

$ virtualenv -p python2.7 pyenv27

(Here python2.7 is required due to src/factorgraph.py was developed in python2.7) Then activate your environment:

$ source path/to/your/evn/pyenv27/bin/activate

and install the requirement file:

$ pip install -r requirements.txt

Experiments

For the sufficient condition check in ER random graph, run

$ python bin/contract_condition.py

For message change along with iteration numbers, to see case of no convergence

$ python bin/converge_rate 0.5 false

For message change along with iteration numbers, to see case with convergence

$ python bin/converge_rate 0.1 true

For application of alpha-bp, run

$ python bin/alpha_compare.py

for experiments without prior.

For application with mmse prior, run

$ python bin/alpha_compare_mmse.py