Simple Scratch Neural Network Library running the XOR problem. The Network goes through 4000 epochs (training loops) while graphing the change in cost. Afterwards a graph is then rendered showing the decision boundary the Network predicts. Base functions for layer math ported from python from NothingButNumpy blog series (below). Matrix function library made by @GreatGameDota . Library inspired by Tensorflow.js This library is very slow with large data sets (especially on Scratch). Links/Resources : Library ported to Javascript : https://github.com/GreatGameDota/neural-network-library Blog post : https://medium.com/towards-artificial-intelligence/nothing-but-numpy-understanding-creating-neural-networks-with-computational-graphs-from-scratch-6299901091b0 Source code : https://github.com/RafayAK/NothingButNumPy #neural #network #ai #algorithm #backpropagation #machine #learning Change Log: ----------------------------------- v0.1 - Initial build: Correct functionality with default weights v0.1.1 - Decrease hard-coded pieces and make it work with any random weights v0.1.2 - Add a dynamic graph for the cost of the training v0.1.3 - Add full prediction analyzer and graph to display it v0.1.4 - Make library multi layer and fully customizable