sscruz/WGAN_MonteCarlo

MC simulations with WGAN

Code to generate MC simulations with WGANs. It includes examples to train a generating network using the MNIST dataset and generator-level Drell--Yan Monte Carlo simulations.

Set up the environment

The environment.yml file contains the packages needed to run the code with pytorch and CUDA 10.2.

conda env create -f environment.yml

conda activate pytorch_v1_cuda_10_2

Examples with different datasets

Generate images using the MNIST dataset

python wgan.py --generator_iters 40000 --model convNNforNist --data mnist --trainingLabel mnisttraining --do_what train --do_what generate

Generate Drell-Yan events using gen-level MC

python wgan.py --generator_iters 100000 --model dense6inputs --data dygen --trainingLabel dytraining --do_what train --do_what generate --n_samples 10000

Package contents

• wgan.py: main script that contains the training algorithm and the parsing of the different options.

• models directory: contains different architectures for the generator and critic networks, that is selected with the --model option. Associated to the critic is the dimensionality and distribution of the latent space, which is also defined here.

• data directory: contains the scripts to handle data. It contains two example classes drellyan_gen and mnist, that are imported through data_loaders. In the context of this repository, data handling includes fetching the data, its preprocessing and its postprocessing, including production of plots.