In this repo are jupyter notebooks used to create three blog posts on computational graphs (one, two and three) and one blog post on python decorators. See the notebooks starting with an underscore for these. The blog posts are the updated version of the notebooks: don't rely on code on the notebook or text to be up to date.

Then there is a module called numpy_autograd that has an implementation of a computational graph. Backprop doesn't work and isn't supported at all at this stage. The __init__.py file converts the notebook in that folder to the script every time it is imported, which isn't the most efficient thing, but then again this isn't a productionised system.