output layers length forgotten after json

[Aedius]

Hello,

when save and charged an nn object with more than one hidden layers, the length of the output layer is forgotten.

i write some mocha test that prove it :

it('trains 2 in 2 out with 1 hidden', function (done) {
    var net = nn()

    //this example should work

    net.train([
        { input: [ 0 , 0.1 ],    output: [ -0.1 , 0.1 ] },
        { input: [ 0.1, 0.1 ],     output: [ 0, 0.2 ] },
        { input: [ 0.2, 0.2 ],     output: [ 0 , 0.4 ] },
        { input: [ 0.3, -0.2 ],     output: [ 0.5 ,  0.1 ] },
        { input: [ 0.4, 0.3 ],     output: [ 0.1 , 0.7 ] },
        { input: [ 0.5, -0.3 ],     output: [ 0.8, 0.2 ] },
        { input: [ 0.6, 0.4 ],     output: [ 0.2, 1 ] },
    ])

    var output = net.send([ 0.5,0.5 ]) // => [ 0, 1 ]

    console.log('trained - , + output for [0.5, 0.5] : [%s]. desiredOutput: [0, 1]', output)

    var json = net.toJson();
    var net = nn()
    net.fromJson(json);
    var output = net.send([ 0.5,0.5 ]) // => [ 0, 1 ]
    console.log('trained - , + output for [0.5, 0.5] : [%s]. desiredOutput: [0, 1]', output)
    assert(output.length = 2)
    done()
})
it('trains 2 in 2 out with 2 hidden', function (done) {

    var net = nn({
        layers: [ 5, 4 ],
        iterations: 2
    })
    //this example should work

    net.train([
        { input: [ 0  ,  0.1 ],     output: [ -0.1 , 0.1 ] },
        { input: [ 0.1,  0.1 ],     output: [  0   , 0.2 ] },
        { input: [ 0.2,  0.2 ],     output: [  0   , 0.4 ] },
        { input: [ 0.3, -0.2 ],     output: [  0.5 , 0.1 ] },
        { input: [ 0.4,  0.3 ],     output: [  0.1 , 0.7 ] },
        { input: [ 0.5, -0.3 ],     output: [  0.8 , 0.2 ] },
        { input: [ 0.6,  0.4 ],     output: [  0.2 , 1 ] },
    ])

    var output = net.send([ 0.5,0.5 ]) // => [ 0, 1 ]

    console.log('trained - , + output for [0.5, 0.5] : [%s]. desiredOutput: [0, 1]', output)

    var json = net.toJson();
    var net = nn()
    net.fromJson(json);
    var output = net.send([ 0.5,0.5 ]) // => [ 0, 1 ]
    console.log('trained - , + output for [0.5, 0.5] : [%s]. desiredOutput: [0, 1]', output)
    assert(output.length == 2)
    done()
})

good work by the way :)

[Hugo-ter-Doest]

The number of layers is not correctly set after reading the Json file. It is always 3 and therefore returns the result of the third layer. Will correct this in a PR.