A simple ANN Wrapper

This framework implements artificial neural networks (ANN) together with some common datasets. It can be used as a package, where you can easily instantiate complex networks and (re-)train them or just run already trained once.

There is a main script, that helps getting used to the framework. Please run

python3 main.py -h

within the src folder to get an overview of this script.

In short, the main script will ensure that

• the output folder is corretly initialized,
• the logger is set up,
• the desired random seed is enforced.

The default values of all command-line parameters are specified in the module configuration.

Example Scripts

There are several example scripts in the folder src/examples, that can be used in combination with the main script.

Here is an example of how to train an MNIST autoencoder:

python3 main.py -r examples.mnist_autoencoder

Or, if running the program on a machine without a graphical interface:

python3 main.py -r examples.mnist_autoencoder -k "{\"allow_plots\": true}"

Implementing new Example Scripts

One can easily write custom example scripts. The only requirement is the existence of a method run(**kwargs). Please checkout existing scripts.

Implemented Datasets

The following datasets are currently supported by the framework.

• MNIST
• CIFAR-10
• CelebA