Simplified neural network

- no external libraries (mostly scratched from brain.js)
- 2 neurons (input - ouput) 1 weight - (in example - see const config)

Description

test2neurns.html

Learns to turn any [0-0.5] input array to [0] and [0.5-1] to [1]
Layers configuration: [1, 1]

Example training data

input:      expected result:
[ 0.0162 ]  [ 0 ]
[ 0.8999 ]  [ 1 ]
[ 0.5707 ]  [ 1 ]
[ 0.4501 ]  [ 0 ]

Example app output (log):

2101 (2) [0.9521130961718607, 2097] (2) [0.9534731391316481, 2098] (2) [0.953440322899051, 2099] (2) [0.9534443265807833, 2100] (2) [0.9534548739868907, 2101]
minimalnn.js:99
[{"val":0.59,"err":0.0059,"bias":-143.4517,"weights":[26.3105]}]
[{"val":0.9534,"err":0.0002,"bias":-11.1668,"weights":[]}]

Mnist example

mnist.html

• Learns to distinguish "8"s, see mnist-database.js / learnNumber