Install and run original GAN using pylearn2 on cuda backend

README.md

Prerequisites

1. Use python2 not python3
2. Suppose the current working directory is $CWD
3. Suppose the data path is $DATA_PATH
4. Suppose the current shell is bash
5. Download following requirements.txt to current working directory
6. Old version of cuda installed(tested on v8.0.44)
7. Old version of cudnn installed(tested on 7.0v4.0)

Install pylearn2

Reference

1. Install dependencies

   pip2 install -qr requirements.txt

2. Clone pylearn2 repo to current working directory

   git clone git://github.com/lisa-lab/pylearn2.git $CWD/pylearn2

3. Change directory

   cd $CWD/pylearn2

4. Install pylearn2

   python2 setup.py develop

5. Run following code to see if pylearn2 is installed correctly

   python2 -c "import pylearn2"

Get dataset

Reference

1. Set your data path here

   echo export PYLEARN2_DATA_PATH=$DATA_PATH >> ~/.bashrc

2. Reload bash profile

   source ~/.bashrc

3. Download possible dataset, using mnist here for example

   python2 $CWD/pylearn2/pylearn2/scripts/datasets/download_mnist.py

Download original GAN implemented by Goodfellow

1. Clone GAN repo to current working directory

   git clone git@github.com:goodfeli/adversarial.git $CWD/adversarial

2. Export adversarial directory to $PYTHONPATH system variable,

   echo export PYTHONPATH=$PYTHONPATH:$CWD/adversarial >> ~/.bashrc

3. Reload bash profile

   source ~/.bashrc

4. Check if the system variable $PYTHONPATH is set correctly

   python2 -c "import adversarial"

Train and test model

1. Train model and the model should be saved in $CWD/adversarial/mnist.pkl

   THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32 python2 -m pylearn2.scripts.train $CWD/adversarial/mnist.yaml

2. Test model

   THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32 python2 -m adversarial.parzen_ll -p $CWD/adversarial/mnist.pkl -d mnist -v -s 0.01

requirements.txt

cycler==0.10.0
Cython==0.28.3
kiwisolver==1.0.1
Mako==1.0.7
MarkupSafe==1.0
matplotlib==2.2.2
nose==1.3.7
numpy==1.14.3
pyparsing==2.2.0
python-dateutil==2.7.3
pytz==2018.4
PyYAML==3.12
scikit-learn==0.19.1
scipy==1.1.0
six==1.11.0
sklearn==0.0
subprocess32==3.5.1
Theano==0.8.0
milk==0.6.1