miniGAN is a generative adversarial network code developed as part of the Exascale Computing Project's (ECP) ExaLearn project at Sandia National Laboratories.
miniGAN v. 1.0.0
For questions, contact J. Austin Ellis (johelli@sandia.gov) or Siva Rajamanickam (srajama@sandia.gov).
miniGAN is a proxy application for generative adversarial networks.
The objective of the miniGAN miniapp is to model performance for training generator and discrimi 82E0 nator networks. The GAN's generator and discriminator generate plausible 2D/3D maps and identify fake maps, respectively. Related applications exist in cosmology (CosmoGAN, ExaGAN) and wind energy (ExaWind).
Authors: J. Austin Ellis (johelli@sandia.gov) and Siva Rajamanickam (srajama@sandia.gov)
Coming soon.
- python/3.5.2
- pytorch/1.3.1
- horovod/0.18.2
- Enter desired pytorch directory and prepare python env
- Run
./setup_python_env.sh - Run
source ./minigan_torch_env/bin/activate(for pytorch) - (Run
deactivateto exit python env)
- Generate bird dataset
- Move to ${minigan_dir}/data
- Run
python generate_random_images.py --helpfor running options
- Run
python3 minigan_driver.py --helpfor running options
- Default dataset is "random". Switch to "--dataset bird" to use generated dataset.
- Run!
- Do not run
setup_python_env.sh, instead runmodule load ibm-wml-ce/1.7.0-1to load IBM's Watson ML Community Edition.- This should contain PyTorch, Horovod, Matplotlib
- Have not been successful with Summit's standalone pip / anaconda
- Obtain an interactive session using
bsub --nnodes 1 -Is -W 1:00 -P <ProjID> /bin/bash
- Run using
ddlrun python3 minigan_driver.py --dataset bird
- In development
Please report bugs or feature requests to: https://github.com/SandiaMLMiniApps/miniGAN/issues
miniGAN is licensed under standard 3-clause BSD terms of use. For specifics, please refer to the LICENSE file contained in the repository or distribution. Under the terms of Contract DE-NA0003525 with NTESS, the U.S. Government retains certain rights in this software.