Skip to content

Latest commit

 

History

History
8 lines (7 loc) · 904 Bytes

README.md

File metadata and controls

8 lines (7 loc) · 904 Bytes

spn-poon-2011

Modification of original implementation of code from Poon,Domingos, Sum-Product Network: a New Deep Architecture, UAI 2011 at http://spn.cs.washington.edu/spn/, for SPN MNIST missing data experiments.

  1. Generate MNIST data for training by running python data/generate_mnist.py (path to raw MNIST data assumes that https://github.com/HUJI-Deep/Generative-ConvACs/blob/master/exp/mnist/generate_mnist.py has already been run).
  2. Follow setup instructions from original code, then compile from code folder using javac -cp .:$MPJ_HOME/lib/mpj.jar -d bin @source.txt
  3. From the bin folder, run mpjrun.sh -np <NUM_PROCS> eval.Run -d M -nsg 1 -ns <NUM_PROCS-1> > ../../results/mnist_run.log where NUM_PROCS is the number of processors available.