git clone https://github.com/BIDData/BIDMach.git
cd BIDMach
git checkout gui
mvn package
First download the models.tar.gz and data.tar.gz from the google drive (https://goo.gl/vqc3rJ)
Extract the models.tar.gz at the BIDMach folder. Extract the data.tar.gz to somewhere you like.
Before running these scripts, config the traindir variable in the scripts as the right data location. Change pretrain_model_dir and pretrain_discriminator_dir if you extract models into different location.
DEMO for mnist:
./bidmach scripts/viz/mnist.ssc
DEMO for CIFAR:
./bidmach scripts/viz/testcifar_norm.ssc
After loading the scripts, type s.launch to start the MCMC process. Use s.stop to stop.
If you want to use discriminator (Require pixel value in [0,256)）, run command o.clipping = true in the shell. And set base as 0 in the UI.
By default, mnist.ssc set clipping as false, testcifar_norm.ssc set clipping as true.
Change L2 weight and discriminator weight to see the effect.