This repo is a Python 3 implementation of a computational graph to do deep learning. I learned in that all big deep learning librairies used that principle so I wanted to understand it and to do that I implemented it.

Simply clone this repo

Essential : Numpy

pip install numpy

Optional : Graphviz

Run fully, fully_mnist, conv or conv_mnist in the terminal

This is the supposed output of the fully file.

This image represent the dataset used, it is a 2d dataset with two classes in a XOR position with a little bit of noise so that perfect separation if impossible

This image represent the graph built

This image represent the exectution of the program in the terminal

This image represent the loss of the graph at each iteration

This image represent the resulting decision boundary

• Numpy - NumPy is the fundamental package for scientific computing with Python
• Graphviz - Graph Visualization Software

• Jean-Gabriel Simard

• Jeremy Fix - CentraleSupélec Teacher