Defaults : By default the ml5.neuralNetwork has simple default architectures for the classification, regression and imageClassificaiton tasks.
default classification layers :
layers:[
{
type: 'dense',
units: this.options.hiddenUnits,
activation: 'relu',
},
{
type: 'dense',
activation: 'softmax',
},
];
default regression layers :
layers: [
{
type: 'dense',
units: this.options.hiddenUnits,
activation: 'relu',
},
{
type: 'dense',
activation: 'sigmoid',
},
];
default imageClassification layers :
layers = [
{
type: 'conv2d',
filters: 8,
kernelSize: 5,
strides: 1,
activation: 'relu',
kernelInitializer: 'varianceScaling',
},
{
type: 'maxPooling2d',
poolSize: [2, 2],
strides: [2, 2],
},
{
type: 'conv2d',
filters: 16,
kernelSize: 5,
strides: 1,
activation: 'relu',
kernelInitializer: 'varianceScaling',
},
{
type: 'maxPooling2d',
poolSize: [2, 2],
strides: [2, 2],
},
{
type: 'flatten',
},
{
type: 'dense',
kernelInitializer: 'varianceScaling',
activation: 'softmax',
},
];
Defining Custom Layers : You can define custom neural network architecture by defining your layers in the options that are passed to the ml5.neuralNetwork on initialization.
A neural network with 3 layers :
const options = {
debug: true,
task: 'classification',
layers: [
{
type: 'dense',
units: 16,
activation: 'relu'
},
{
type: 'dense',
units: 16,
activation: 'sigmoid'
},
{
type: 'dense',
activation: 'sigmoid'
}
]
};
const nn = ml5.neuralNetwork(options);
