MCLV-RBM

Training RBMs using MCLV

Setting up the code

You will need Anaconda Create the working directory and navigate into it

unzip MCLV.zip

You'll see the following folder structure

LICENSE  MCLV.zip  py  README.MD  results

Creating the required conda environment

conda create -n mclv_env python=3.6 pip

source activate mclv_env

pip install --upgrade pip

pip install numpy

pip install pandas

pip install scipy

pip install matplotlib

pip install sklearn

pip install seaborn

For pytorch please refer to http://pytorch.org/ For our system we did:

pip install http://download.pytorch.org/whl/cu80/torch-0.2.0.post3-cp36-cp36m-manylinux1_x86_64.whl 

pip install torchvision

To activate the required conda environment

source activate mclv_env

To run the program the following commands can be used

To display options

python3 py/main.py -h

To Setup Data(Requires Tensorflow or email us for the processed dataset if you can't install tribeflow):

python3 py/setup_data.py -b `pwd`/ 

For PCD Results

python3 py/main.py -b `pwd`/ -n 1 --method PCD -cdk 10 -tot 100 --plateau 10 --hidden 32 --final-likelihood --filename h32pcd1_3

For CD Results

python3 py/main.py -b `pwd`/ -n 1 --method CD -cdk 10 -tot 100 --plateau 10 --hidden 32 --final-likelihood --filename h32pcd1_3

For MCLV Results

python3 py/main.py -b `pwd`/ -n 1 --method MCLV -mclvk 10 -cdk 10 -wm 15 -tot 100 --plateau 10 --hidden 32 --final-likelihood --filename h32pcd1_3

Notes

You'll need a GPU with at least 8GB free memory, if not use the GPU-LIMIT to tune the parallelism for the likelihood computation. We didn't implement bloom filters at the present but used python sets instead.