NeuralNetwork

You can create a NeuralNetwork object with Engine.createNeuralNetwork("String id") . Please take a look at the NeuralNetwork Sine Generator Snippet to take a look at a working example.

const var nn = Engine.createNeuralNetwork("NeuralNetwork");

Class methods

build

Create a network using the given JSON for the layer setup. Edit on GitHub

NeuralNetwork.build( var modelJSON)

clearModel

Destroys the model and allows rebuilding using a different layout JSON. Edit on GitHub

NeuralNetwork.clearModel()

connectToGlobalCables

Connects the network to a input and output global cable. Edit on GitHub

NeuralNetwork.connectToGlobalCables(String inputId, String outputId)

createModelJSONFromTextFile

Helper function to create a JSON model definition from the Pytorch print(model) output. Edit on GitHub

NeuralNetwork.createModelJSONFromTextFile(var fileObject)

getModelJSON

Returns the model JSON. Edit on GitHub

NeuralNetwork.getModelJSON()

loadOnnxModel

Loads the ONNX runtime model for spectral analysis. Edit on GitHub

NeuralNetwork.loadOnnxModel( var base64Data, int numOutputValues)

loadPytorchModel

Loads the model layout and weights from a Pytorch model JSON. Edit on GitHub

NeuralNetwork.loadPytorchModel( var modelJSON)

loadTensorFlowModel

Loads the model layout and weights from a tensorflow model JSON. Edit on GitHub

NeuralNetwork.loadTensorFlowModel( var modelJSON)

loadWeights

Loads the weights from the JSON object. Edit on GitHub

NeuralNetwork.loadWeights( var weightData)

process

Runs inference on the given input and returns either a single float or a reference to the output buffer. Edit on GitHub

NeuralNetwork.process(var input)

processFFTSpectrum

Processes the FFT spectrum and returns the output tensor as array of float numbers. Edit on GitHub

NeuralNetwork.processFFTSpectrum(var fftObject, int numFreqPixels, int numTimePixels)

reset

Resets the network pipeline. Edit on GitHub

NeuralNetwork.reset()