A TensorFlow-inspired ANN library in pure ES6 without any external dependencies.
The main library. Implemented Linear, Sigmoid and MSE nodes.
ReLU and Softmax activation, as well as Cross Entropy cost function, are planned.
The end-goal here is to have a network that decently performs on the MNIST dataset.
Some necessary minimalistic maths stuff. Mostly linear algebra.
An end-to-end example of deploying the Ann to work out the Boston Housing Prices.
$ yarn
$ ./node_modules/.bin/jest .
$ node --experimental-modules boston_demo.mjs