JavaSript implementation of feedforward neural network with backpropagarion algorithm.
This project is still in development phase. Further development should bring extended description, add more use cases and involve code refactoring and optimalisation.
P.S. Sorry for my english. If You wish to help me with this project or correct my english description - You are welcome :)
- node js - Basically this is not strong requirement. There is no problem to run this library in modern browser. I decide to use node.js because there is more CPU power to use by JavaScript
Run this command in the project main directory:
node demo/xor.js
Example output:
learn in epoch 1647 error 0.007963763295449296
test net. epoch 1647 error 0.007963763295449296
[ 0, 0 ] 0.06878616562023242
[ 0, 1 ] 0.9628854533576559
[ 1, 0 ] 0.9301744412288724
[ 1, 1 ] 0.06920811758028442
Objective: learn neural network of simple number sequence and use this net to "predict" next number in test sequence
Run this command in the project main directory:
node demo/sequence-prediction.js
learn in epoch 3334 error 0.00009983697113540062
test net. epoch 3334 error 0.00009983697113540062
number sequence: 1, 2, 3, 4, 3, 2, 1, 2, 3, 4
input: [ 4, 3, 2 ] "predicted" output: round(1), raw(1.04923740569591) normalized input: [ 0.8, 0.6, 0.4 ] normalized output: 0.209847481139182
input: [ 3, 2, 1 ] "predicted" output: round(2), raw(1.7728240967319415) normalized input: [ 0.6, 0.4, 0.2 ] normalized output: 0.3545648193463883