Quick start

To get running on any supported platform, do the following in a clean shell:

Set up package

$ git clone [email protected]:asogaard/adversarial.git
$ cd adversarial
$ source install.sh
$ source activate.sh
$ pip install -e .  # To be confirmed; may need to be done in each conda environment.

This installs the supported conda environments and activates the one for CPU running.

Stage some data

$ source scripts/get_data.sh

If run elsewhere than lxplus, this will download a 1.4GB HDF5 data file.

Test run

$ python -m run.adversarial.train --help
$ python -m run.adversarial.train --train

This shows the supported arguments to the run/adversarial/train.py script, and starts training using the TensorFlow backend. Several scripts — for training, testing, and optimisation — are available in the run/ directory and can be run either as modules or as scripts. Tab-completion will be enabled for all of them.