Learning Markov Random Fields for Combinatorial Structures via Sampling through Lov'asz Local Lemma

full paper: https://arxiv.org/abs/2212.00296

Requirements

the code needs to install pytroch

pip install torch

Datasets

to generate the Rand-k-cnf dataset we need

pip install cnfgen
pip install 'python-sat[pblib,aiger]'

1. Uniform sampling comparison

sampling SAT-CNF solutions uniformly.

./run_weighted_random_ksat.py lll 5 300 300 uniform
./run_weighted_random_ksat.py lll 5 300 300 weighted

[nelson, spur, unigen, quciksampler, cmsgen, bddsampler, kus, smarch, searchtreesampler]

note that, for unigen, cmsgen you need to goto its repository and install the program according to the readme.md file.

To run the kus program, you need to install pydot.

All the baseline methods included for comparison are cloned from its public-available repositories. If you want to run the rest programs, you need to goto every link and install it.

• Spur
• Unigen
• QuickSampler
• CMSGen
• KUS.
• Smarch
• STS

2. Weighted sampling comparison

Requirement:

pip install waps