Weird results

[clive-g-brown]

Description

When running the vanilla xor, I get widly varying results after training, yet with a very low measured error. almost random results with low error, both in debugger and when printing to console.

Screenshots

5 consecutive runs (error is line 1):

-0
[ 1 ]
[ 1 ]
[ -1 ]
[ 1 ]

-0
[ 0 ]
[ 1 ]
[ 0 ]
[ 0 ]

0.00019156340528790638
[ 0.5116388521017431 ]
[ 0.3428954069380361 ]
[ 0.5116388521017431 ]
[ 0.3428954069380361 ]

-0
[ 0.34173701088298947 ]
[ 0.005307673690185314 ]
[ 0.0017903267041861595 ]
[ 6.842180212331013e-7 ]

-0
[ 1 ]
[ 0 ]
[ 0 ]
[ 0 ]

Files

A list of relevant files for this issue. This will help people navigate the project and offer some clues of where to start.

To Reproduce

const chai = require('chai')
const assert = chai.assert

let { Network, methods } = require('@liquid-carrot/carrot');

// this network learns the XOR gate (through neuro-evolution)
async function execute () {
// no hidden layers...
var network = new Network(2,1);

// XOR dataset
var trainingSet = [
{ input: [0,0], output: [0] },
{ input: [0,1], output: [1] },
{ input: [1,0], output: [1] },
{ input: [1,1], output: [0] }
];

const results = await network.evolve(trainingSet, {
mutation: methods.mutation.FFW,
equal: true,
error: 0.0003,
elitism: 10,
mutation_rate: 0.3,
threads: 5
});
console.log(results.error)
assert.isBelow(results.error, 0.0003);
// and it works!
console.log(network.activate([0,0])); // 0.2413
console.log(network.activate([0,1])); // 1.0000
console.log(network.activate([1,0])); // 0.7663
console.log(network.activate([1,1])); // 0.008
}

execute();

/// threads 1, 1000 iterations -- logging

teration 100 fitness -0.0011220021193443105 error 0.0005220021193443104
iteration 200 fitness -0.0006363003302362967 error 0.0000363003302362966
iteration 300 fitness -0.0006277835038999049 error 0.00002778350389990485
iteration 400 fitness -0.0006250023524746145 error 0.00002500235247461448
iteration 500 fitness -0.0006244568900950992 error 0.000024456890095099146
iteration 600 fitness -0.0006058503711863999 error 0.00000585037118639982
iteration 700 fitness -0.0006041840989964885 error 0.000004184098996488483
iteration 800 fitness -0.0006020429427010222 error 0.0000020429427010221213
iteration 900 fitness -0.0006014404264906937 error 0.000001440426490693686
iteration 1000 fitness -0.0006014404264906937 error 0.000001440426490693686
0.000001440426490693686
[ 0.99596362676882 ]
[ 0.496402339649377 ]
[ 0.41957367383984256 ]
[ 0.034813165252755914 ]
neon_carrot % node test1.js
iteration 100 fitness -0.0012000000000000001 error -0
iteration 200 fitness -0.0012000000000000001 error -0
iteration 300 fitness -0.0012000000000000001 error -0
iteration 400 fitness -0.0012000000000000001 error -0
iteration 500 fitness -0.0012000000000000001 error -0
iteration 600 fitness -0.0012000000000000001 error -0
iteration 700 fitness -0.0011 error -0
iteration 800 fitness -0.0011 error -0
iteration 900 fitness -0.0011 error -0
iteration 1000 fitness -0.0011 error -0
-0
[ 0.27516382951966845 ]
[ 0.3855840206635007 ]
[ 0.27516382951966845 ]
[ 0.3855840206635007 ]
neon_carrot % node test1.js
iteration 100 fitness -0.0006000000000000001 error -0
iteration 200 fitness -0.0006000000000000001 error -0
iteration 300 fitness -0.0006000000000000001 error -0
iteration 400 fitness -0.0006000000000000001 error -0
iteration 500 fitness -0.0006000000000000001 error -0
iteration 600 fitness -0.0006000000000000001 error -0
iteration 700 fitness -0.0006000000000000001 error -0
iteration 800 fitness -0.0006000000000000001 error -0
iteration 900 fitness -0.0006000000000000001 error -0
iteration 1000 fitness -0.0006000000000000001 error -0
-0
[ 1 ]
[ -1 ]
[ 1 ]
[ 1 ]

// then after many tries it gets it right

iteration 100 fitness -0.0004567252593608716 error 0.0002567252593608716
iteration 200 fitness -0.00021686014209889781 error 0.000016860142098897805
iteration 300 fitness -0.00020795225452568586 error 0.000007952254525685855
iteration 400 fitness -0.0002029582806041947 error 0.0000029582806041947003
iteration 500 fitness -0.0002029582806041947 error 0.0000029582806041947003
iteration 600 fitness -0.00020277357041923415 error 0.0000027735704192341406
iteration 700 fitness -0.00020082334109185242 error 8.233410918524102e-7
iteration 800 fitness -0.00020082334109185242 error 8.233410918524102e-7
iteration 900 fitness -0.00020082334109185242 error 8.233410918524102e-7
iteration 1000 fitness -0.00020082334109185242 error 8.233410918524102e-7
8.233410918524102e-7
[ 0.0002002736795891913 ]
[ 0.9990157960831416 ]
[ 1.0006023497616376 ]
[ 0.0013862799989068364 ]

[clive-g-brown]

Hi all. I tried this again, using the example given in the docs (same as above), just with console outputs. Its still generating basically random outputs, only occasionally right.