An interactive neural net demo written in vanilla javascript meant to foster intuitive understanding of neural net training.
Short for "neural nets in javascript" and pronounced "ninjas", because duh!
Hit the "play" button () below to get started!
Add data by clicking or right-clicking on the output (bottom-right) graph.
Select a neuron in the graph to plot its output in the top-right graph.
# hidden layers # neurons
Selected neuron
Network output

Not snappy?

This is not fast, performant code by any means. We're not even vetorizing operations, if that means anything to you. Instead, the code prioritizes being approachable, readable, and instructive. Check it out and tinker!

Bakclog, AKA Coming soon(ish):

  • Pin & Force parameters
  • Visualization of validation data
  • Graphs of selected parameters' histroy
  • Activation function selection and visualization
  • Visualization of learning slowdown, RELU death, etc.
I'm just here for the padding, man.