Sympyle

Simple Symbolic Graphs in Python

About

Sympyle is a Python library to demonstrate the inner workings of Computational Graphs. Computational Graphs are used by highly optimised computational frameworks like Tensorflow and Pytorch.

However, these frameworks make several assumptions and optimisations in order to optimise for speed and memory. This often makes it harder to understand the inner workings of how these libraries work.

What Sympyle is

Sympyle is a simplified model library to demonstrate the working of computational graphs, and how backpropagation works on arbitrary ‘networks’.

Each “operation” will contain detailed documentation on the derivation, and whenever possible, additional resources to better understand it.

What Sympyle is Not

Sympyle is not a production ready library. It should not be thought of as a competitor to larger frameworks like Tensorflow.

It will also likely not implement all the ops that other libraries use.