Predicting Satisfiability via End-to-End Learning
See here of paper: https://www.cs.ubc.ca/labs/algorithms/Projects/End2EndSAT/paper.pdf
Download repository and change directories:
git clone https://github.com/ChrisCameron1/End2EndSAT.git && cd End2EndSAT
Download data:
wget https://www.cs.ubc.ca/labs/algorithms/Projects/End2EndSAT/data.zip && unzip data.zip
We recommend to use with CometML for monitoring experiments. Please create a free cometML account at https://www.comet.com/site/. Open the mypaths.py file and replace with you username and key.
COMET={'un': '[your_username]',
'key': '[your_key]'}
Each directory in data/ corresponds to a different dataset. To run with the hyper paramters from the paper on the 100 variable dataset, first untar that dataset (will take a while) and create corresponding logs directory:
tar -xvzf data/100.tar.gz && mv 100 data/ && mkdir logs && mkdir logs/100
and run:
python train.py 100 -m nn_raw -ps -le 8 -lff 2 -u 128 -v
Can add the -cn parameter for use without CometML but we do no support proper experimental logging in this setting.