Introduction

• GANumpy is a Generative Adversarial Network written in pure Numpy (educational purpose only).
• It uses Yaae, my custom automatic differentiation engine also written in pure Numpy.
• The GAN was trained on MNIST dataset. To speed up the training time, only a subset of digit samples (1,2,3) were used. Here are the results:

Installation

• Create a virtual environment in the root folder using virtualenv and activate it.

# On Linux terminal, using virtualenv.
virtualenv my_ganumpy_env
# Activate it.
source my_ganumpy_env/bin/activate

• Install requirements.txt.

pip install -r requirements.txt
# Tidy up the root folder.
python3 setup.py clean