Source code for "Towards a Deeper Understanding of Adversarial Losses"

Below we assume the working directory is the repository root.

• Using pipenv (recommended)

  Make sure pipenv is installed. (If not, simply run pip install --user pipenv.)

  # Install the dependencies
  pipenv install
  # Activate the virtual environment
  pipenv shell

• Using pip

  # Install the dependencies
  pip install -r requirements.txt

# Download the training data
./scripts/download_data.sh
# Store the training data to shared memory
./scripts/process_data.sh

You can also download the MNIST handwritten digit database manually here.

We provide several shell scripts for easy managing the experiments. (See scripts/README.md for a detailed documentation.)

Below we assume the working directory is the repository root.

1. Run the following command to set up a new experiment with default settings.

   # Set up a new experiment (for one run only)
   ./scripts/setup_exp.sh -r 1 "./exp/my_experiment/"

2. Modify the configuration files for different experimental settings. The configuration file can be found at ./exp/my_experiment/config.yaml.

3. Train the model by running the following command.

   # Train the model (on GPU 0)
   ./scripts/run_train.sh -c -g 0 "./exp/my_experiment/"

For each run, there will be three folders created in the experiment folder.

• logs/: contain all the logs created
• model/: contain the trained model
• src/: contain a backup of the source code

Note that the validation results can be found in the logs/ folder.