394 lines
13 KiB
JavaScript
394 lines
13 KiB
JavaScript
|
/* Import */
|
||
|
|
var chai = require('chai');
|
||
|
|
var assert = chai.assert;
|
||
|
|
|
||
|
|
/* Shorten var names */
|
||
|
|
var { architect, Network, methods, config } = neataptic;
|
||
|
|
|
||
|
|
/* Turn off warnings */
|
||
|
|
config.warnings = false;
|
||
|
|
|
||
|
|
/* Functions used in the testing process */
|
||
|
|
function checkMutation (method) {
|
||
|
|
var network = new architect.Perceptron(2, 4, 4, 4, 2);
|
||
|
|
network.mutate(methods.mutation.ADD_GATE);
|
||
|
|
network.mutate(methods.mutation.ADD_BACK_CONN);
|
||
|
|
network.mutate(methods.mutation.ADD_SELF_CONN);
|
||
|
|
|
||
|
|
var originalOutput = [];
|
||
|
|
var i, j;
|
||
|
|
for (i = 0; i <= 10; i++) {
|
||
|
|
for (j = 0; j <= 10; j++) {
|
||
|
|
originalOutput.push(network.activate([i / 10, j / 10]));
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
network.mutate(method);
|
||
|
|
|
||
|
|
var mutatedOutput = [];
|
||
|
|
|
||
|
|
for (i = 0; i <= 10; i++) {
|
||
|
|
for (j = 0; j <= 10; j++) {
|
||
|
|
mutatedOutput.push(network.activate([i / 10, j / 10]));
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
assert.notDeepEqual(originalOutput, mutatedOutput, 'Output of original network should be different from the mutated network!');
|
||
|
|
}
|
||
|
|
|
||
|
|
function learnSet (set, iterations, error) {
|
||
|
|
var network = new architect.Perceptron(set[0].input.length, 5, set[0].output.length);
|
||
|
|
|
||
|
|
var options = {
|
||
|
|
iterations: iterations,
|
||
|
|
error: error,
|
||
|
|
shuffle: true,
|
||
|
|
rate: 0.3,
|
||
|
|
momentum: 0.9
|
||
|
|
};
|
||
|
|
|
||
|
|
var results = network.train(set, options);
|
||
|
|
|
||
|
|
assert.isBelow(results.error, error);
|
||
|
|
}
|
||
|
|
|
||
|
|
function testEquality (original, copied) {
|
||
|
|
for (var j = 0; j < 50; j++) {
|
||
|
|
var input = [];
|
||
|
|
var a;
|
||
|
|
for (a = 0; a < original.input; a++) {
|
||
|
|
input.push(Math.random());
|
||
|
|
}
|
||
|
|
|
||
|
|
var ORout = original.activate([input]);
|
||
|
|
var COout = copied instanceof Network ? copied.activate([input]) : copied([input]);
|
||
|
|
|
||
|
|
for (a = 0; a < original.output; a++) {
|
||
|
|
ORout[a] = ORout[a].toFixed(9);
|
||
|
|
COout[a] = COout[a].toFixed(9);
|
||
|
|
}
|
||
|
|
assert.deepEqual(ORout, COout, copied instanceof Network
|
||
|
|
? 'Original and JSON copied networks are not the same!'
|
||
|
|
: 'Original and standalone networks are not the same!'
|
||
|
|
);
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
/*******************************************************************************************
|
||
|
|
Test the performance of networks
|
||
|
|
*******************************************************************************************/
|
||
|
|
|
||
|
|
describe('Networks', function () {
|
||
|
|
describe('Mutation', function () {
|
||
|
|
it('ADD_NODE', function () {
|
||
|
|
checkMutation(methods.mutation.ADD_NODE);
|
||
|
|
});
|
||
|
|
it('ADD_CONNECTION', function () {
|
||
|
|
checkMutation(methods.mutation.ADD_CONN);
|
||
|
|
});
|
||
|
|
it('MOD_BIAS', function () {
|
||
|
|
checkMutation(methods.mutation.MOD_BIAS);
|
||
|
|
});
|
||
|
|
it('MOD_WEIGHT', function () {
|
||
|
|
checkMutation(methods.mutation.MOD_WEIGHT);
|
||
|
|
});
|
||
|
|
it('SUB_CONN', function () {
|
||
|
|
checkMutation(methods.mutation.SUB_CONN);
|
||
|
|
});
|
||
|
|
it('SUB_NODE', function () {
|
||
|
|
checkMutation(methods.mutation.SUB_NODE);
|
||
|
|
});
|
||
|
|
it('MOD_ACTIVATION', function () {
|
||
|
|
checkMutation(methods.mutation.MOD_ACTIVATION);
|
||
|
|
});
|
||
|
|
it('ADD_GATE', function () {
|
||
|
|
checkMutation(methods.mutation.ADD_GATE);
|
||
|
|
});
|
||
|
|
it('SUB_GATE', function () {
|
||
|
|
checkMutation(methods.mutation.SUB_GATE);
|
||
|
|
});
|
||
|
|
it('ADD_SELF_CONN', function () {
|
||
|
|
checkMutation(methods.mutation.ADD_SELF_CONN);
|
||
|
|
});
|
||
|
|
it('SUB_SELF_CONN', function () {
|
||
|
|
checkMutation(methods.mutation.SUB_SELF_CONN);
|
||
|
|
});
|
||
|
|
it('ADD_BACK_CONN', function () {
|
||
|
|
checkMutation(methods.mutation.ADD_BACK_CONN);
|
||
|
|
});
|
||
|
|
it('SUB_BACK_CONN', function () {
|
||
|
|
checkMutation(methods.mutation.SUB_BACK_CONN);
|
||
|
|
});
|
||
|
|
it('SWAP_NODES', function () {
|
||
|
|
checkMutation(methods.mutation.SWAP_NODES);
|
||
|
|
});
|
||
|
|
});
|
||
|
|
describe('Structure', function () {
|
||
|
|
it('Feed-forward', function () {
|
||
|
|
this.timeout(30000);
|
||
|
|
var network1 = new Network(2, 2);
|
||
|
|
var network2 = new Network(2, 2);
|
||
|
|
|
||
|
|
// mutate it a couple of times
|
||
|
|
var i;
|
||
|
|
for (i = 0; i < 100; i++) {
|
||
|
|
network1.mutate(methods.mutation.ADD_NODE);
|
||
|
|
network2.mutate(methods.mutation.ADD_NODE);
|
||
|
|
}
|
||
|
|
for (i = 0; i < 400; i++) {
|
||
|
|
network1.mutate(methods.mutation.ADD_CONN);
|
||
|
|
network2.mutate(methods.mutation.ADD_NODE);
|
||
|
|
}
|
||
|
|
|
||
|
|
// Crossover
|
||
|
|
var network = Network.crossOver(network1, network2);
|
||
|
|
|
||
|
|
// Check if the network is feed-forward correctly
|
||
|
|
for (i = 0; i < network.connections.length; i++) {
|
||
|
|
let from = network.nodes.indexOf(network.connections[i].from);
|
||
|
|
let to = network.nodes.indexOf(network.connections[i].to);
|
||
|
|
|
||
|
|
// Exception will be made for memory connections soon
|
||
|
|
assert.isBelow(from, to, 'network is not feeding forward correctly');
|
||
|
|
}
|
||
|
|
});
|
||
|
|
it('from/toJSON equivalency', function () {
|
||
|
|
this.timeout(10000);
|
||
|
|
var original, copy;
|
||
|
|
original = new architect.Perceptron(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
copy = Network.fromJSON(original.toJSON());
|
||
|
|
testEquality(original, copy);
|
||
|
|
|
||
|
|
original = new Network(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
copy = Network.fromJSON(original.toJSON());
|
||
|
|
testEquality(original, copy);
|
||
|
|
|
||
|
|
original = new architect.LSTM(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
copy = Network.fromJSON(original.toJSON());
|
||
|
|
testEquality(original, copy);
|
||
|
|
|
||
|
|
original = new architect.GRU(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
copy = Network.fromJSON(original.toJSON());
|
||
|
|
testEquality(original, copy);
|
||
|
|
|
||
|
|
original = new architect.Random(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 10 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
copy = Network.fromJSON(original.toJSON());
|
||
|
|
testEquality(original, copy);
|
||
|
|
|
||
|
|
original = new architect.NARX(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 10 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
copy = Network.fromJSON(original.toJSON());
|
||
|
|
testEquality(original, copy);
|
||
|
|
|
||
|
|
original = new architect.Hopfield(Math.floor(Math.random() * 5 + 1));
|
||
|
|
copy = Network.fromJSON(original.toJSON());
|
||
|
|
testEquality(original, copy);
|
||
|
|
});
|
||
|
|
it('standalone equivalency', function () {
|
||
|
|
this.timeout(10000);
|
||
|
|
var original;
|
||
|
|
original = new architect.Perceptron(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
eval(original.standalone());
|
||
|
|
testEquality(original, activate);
|
||
|
|
|
||
|
|
original = new Network(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
eval(original.standalone());
|
||
|
|
testEquality(original, activate);
|
||
|
|
|
||
|
|
original = new architect.LSTM(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
eval(original.standalone());
|
||
|
|
testEquality(original, activate);
|
||
|
|
|
||
|
|
original = new architect.GRU(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
eval(original.standalone());
|
||
|
|
testEquality(original, activate);
|
||
|
|
|
||
|
|
original = new architect.Random(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 10 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
eval(original.standalone());
|
||
|
|
testEquality(original, activate);
|
||
|
|
|
||
|
|
original = new architect.NARX(Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1), Math.floor(Math.random() * 5 + 1));
|
||
|
|
eval(original.standalone());
|
||
|
|
testEquality(original, activate);
|
||
|
|
|
||
|
|
original = new architect.Hopfield(Math.floor(Math.random() * 5 + 1));
|
||
|
|
eval(original.standalone());
|
||
|
|
testEquality(original, activate);
|
||
|
|
});
|
||
|
|
});
|
||
|
|
describe('Learning capability', function () {
|
||
|
|
it('AND gate', function () {
|
||
|
|
learnSet([
|
||
|
|
{ input: [0, 0], output: [0] },
|
||
|
|
{ input: [0, 1], output: [0] },
|
||
|
|
{ input: [1, 0], output: [0] },
|
||
|
|
{ input: [1, 1], output: [1] }
|
||
|
|
], 1000, 0.002);
|
||
|
|
});
|
||
|
|
it('XOR gate', function () {
|
||
|
|
learnSet([
|
||
|
|
{ input: [0, 0], output: [0] },
|
||
|
|
{ input: [0, 1], output: [1] },
|
||
|
|
{ input: [1, 0], output: [1] },
|
||
|
|
{ input: [1, 1], output: [0] }
|
||
|
|
], 3000, 0.002);
|
||
|
|
});
|
||
|
|
it('NOT gate', function () {
|
||
|
|
learnSet([
|
||
|
|
{ input: [0], output: [1] },
|
||
|
|
{ input: [1], output: [0] }
|
||
|
|
], 1000, 0.002);
|
||
|
|
});
|
||
|
|
it('XNOR gate', function () {
|
||
|
|
learnSet([
|
||
|
|
{ input: [0, 0], output: [1] },
|
||
|
|
{ input: [0, 1], output: [0] },
|
||
|
|
{ input: [1, 0], output: [0] },
|
||
|
|
{ input: [1, 1], output: [1] }
|
||
|
|
], 3000, 0.002);
|
||
|
|
});
|
||
|
|
it('OR gate', function () {
|
||
|
|
learnSet([
|
||
|
|
{ input: [0, 0], output: [0] },
|
||
|
|
{ input: [0, 1], output: [1] },
|
||
|
|
{ input: [1, 0], output: [1] },
|
||
|
|
{ input: [1, 1], output: [1] }
|
||
|
|
], 1000, 0.002);
|
||
|
|
});
|
||
|
|
it('SIN function', function () {
|
||
|
|
this.timeout(30000);
|
||
|
|
var set = [];
|
||
|
|
|
||
|
|
while (set.length < 100) {
|
||
|
|
var inputValue = Math.random() * Math.PI * 2;
|
||
|
|
set.push({
|
||
|
|
input: [inputValue / (Math.PI * 2)],
|
||
|
|
output: [(Math.sin(inputValue) + 1) / 2]
|
||
|
|
});
|
||
|
|
}
|
||
|
|
|
||
|
|
learnSet(set, 1000, 0.05);
|
||
|
|
});
|
||
|
|
it('Bigger than', function () {
|
||
|
|
this.timeout(30000);
|
||
|
|
var set = [];
|
||
|
|
|
||
|
|
for (var i = 0; i < 100; i++) {
|
||
|
|
var x = Math.random();
|
||
|
|
var y = Math.random();
|
||
|
|
var z = x > y ? 1 : 0;
|
||
|
|
|
||
|
|
set.push({ input: [x, y], output: [z] });
|
||
|
|
}
|
||
|
|
|
||
|
|
learnSet(set, 500, 0.05);
|
||
|
|
});
|
||
|
|
it('LSTM XOR', function () {
|
||
|
|
this.timeout(30000);
|
||
|
|
var lstm = new architect.LSTM(1, 1, 1);
|
||
|
|
|
||
|
|
lstm.train([
|
||
|
|
{ input: [0], output: [0] },
|
||
|
|
{ input: [1], output: [1] },
|
||
|
|
{ input: [1], output: [0] },
|
||
|
|
{ input: [0], output: [1] },
|
||
|
|
{ input: [0], output: [0] }
|
||
|
|
], {
|
||
|
|
error: 0.001,
|
||
|
|
iterations: 5000,
|
||
|
|
rate: 0.3
|
||
|
|
});
|
||
|
|
|
||
|
|
lstm.activate([0]);
|
||
|
|
|
||
|
|
function getActivation(sensors) {
|
||
|
|
return lstm.activate(sensors)[0];
|
||
|
|
}
|
||
|
|
|
||
|
|
assert.isBelow(0.9, getActivation([1]), 'LSTM error');
|
||
|
|
assert.isBelow(getActivation([1]), 0.1, 'LSTM error');
|
||
|
|
assert.isBelow(0.9, getActivation([0]), 'LSTM error');
|
||
|
|
assert.isBelow(getActivation([0]), 0.1, 'LSTM error');
|
||
|
|
});
|
||
|
|
it('GRU XOR', function () {
|
||
|
|
this.timeout(30000);
|
||
|
|
var gru = new architect.GRU(1, 2, 1);
|
||
|
|
|
||
|
|
gru.train([
|
||
|
|
{ input: [0], output: [0] },
|
||
|
|
{ input: [1], output: [1] },
|
||
|
|
{ input: [1], output: [0] },
|
||
|
|
{ input: [0], output: [1] },
|
||
|
|
{ input: [0], output: [0] }
|
||
|
|
], {
|
||
|
|
error: 0.001,
|
||
|
|
iterations: 5000,
|
||
|
|
rate: 0.1,
|
||
|
|
clear: true
|
||
|
|
});
|
||
|
|
|
||
|
|
gru.activate([0]);
|
||
|
|
|
||
|
|
function getActivation(sensors) {
|
||
|
|
return gru.activate(sensors)[0];
|
||
|
|
}
|
||
|
|
|
||
|
|
assert.isBelow(0.9, getActivation([1]), 'GRU error');
|
||
|
|
assert.isBelow(getActivation([1]), 0.1, 'GRU error');
|
||
|
|
assert.isBelow(0.9, getActivation([0]), 'GRU error');
|
||
|
|
assert.isBelow(getActivation([0]), 0.1, 'GRU error');
|
||
|
|
});
|
||
|
|
it('NARX Sequence', function () {
|
||
|
|
var narx = new architect.NARX(1, 5, 1, 3, 3);
|
||
|
|
|
||
|
|
// Train the XOR gate (in sequence!)
|
||
|
|
var trainingData = [
|
||
|
|
{ input: [0], output: [0] },
|
||
|
|
{ input: [0], output: [0] },
|
||
|
|
{ input: [0], output: [1] },
|
||
|
|
{ input: [1], output: [0] },
|
||
|
|
{ input: [0], output: [0] },
|
||
|
|
{ input: [0], output: [0] },
|
||
|
|
{ input: [0], output: [1] }
|
||
|
|
];
|
||
|
|
|
||
|
|
narx.train(trainingData, {
|
||
|
|
iterations: 1000,
|
||
|
|
error: 0.005,
|
||
|
|
rate: 0.05
|
||
|
|
});
|
||
|
|
|
||
|
|
assert.isBelow(narx.test(trainingData).error, 0.005);
|
||
|
|
});
|
||
|
|
it('SIN + COS', function () {
|
||
|
|
this.timeout(30000);
|
||
|
|
var set = [];
|
||
|
|
|
||
|
|
while (set.length < 100) {
|
||
|
|
var inputValue = Math.random() * Math.PI * 2;
|
||
|
|
set.push({
|
||
|
|
input: [inputValue / (Math.PI * 2)],
|
||
|
|
output: [
|
||
|
|
(Math.sin(inputValue) + 1) / 2,
|
||
|
|
(Math.cos(inputValue) + 1) / 2
|
||
|
|
]
|
||
|
|
});
|
||
|
|
}
|
||
|
|
|
||
|
|
learnSet(set, 1000, 0.05);
|
||
|
|
});
|
||
|
|
it('SHIFT', function () {
|
||
|
|
var set = [];
|
||
|
|
|
||
|
|
for (var i = 0; i < 1000; i++) {
|
||
|
|
var x = Math.random();
|
||
|
|
var y = Math.random();
|
||
|
|
var z = Math.random();
|
||
|
|
|
||
|
|
set.push({ input: [x, y, z], output: [z, x, y] });
|
||
|
|
}
|
||
|
|
|
||
|
|
learnSet(set, 500, 0.03);
|
||
|
|
});
|
||
|
|
});
|
||
|
|
});
|