Kiss.Network Class Reference

This is the main class of kissinference, it allows you to load and use Kiss neural neturoks. More...

Inherits System.IDisposable.

Collaboration diagram for Kiss.Network:

Public Member Functions
delegate void	ResultDlg (float[] result, Network network)
	Network (string path, bool verbose=false)
	Constructs a new Network object by loading a network off disk.
AutoResetEvent	Run (Spectra spectra, ResultDlg callback)
	This method runs an inference pass on the given spectra asynchronously.
AutoResetEvent	Run (float[] data, ResultDlg callback)
	This method runs an inference pass on the given spectra asynchronously.
void	Dispose ()

Protected Member Functions
virtual void	Dispose (bool disposing)

Properties
int	InputSize [get]
	This Property holds the number of input values the loaded network expects.
int	OutputSize [get]
	This Property holds the number of output values the loaded network expects.
string	Purpose [get]
	This Property holds a string describeing the purpose the network was trained on.
string	InputLabel [get]
	This Property holds a string describeing the type of input the network expects.
bool	ComplexInput [get]
	This Property is true if the network expects a complex valued input.
string[]	OutputLabels [get]
	This Property is an array of OutputSize with strings nameing what every output corrisponds to.
bool[]	OutputMask [get, set]
	This Property is an array of bools enableing or disableing a given output.

Detailed Description

This is the main class of kissinference, it allows you to load and use Kiss neural neturoks.

Constructor & Destructor Documentation

Network()

Kiss.Network.Network ( string path, bool verbose = false )

inline

Constructs a new Network object by loading a network off disk.

Parameters

path	the path to the onnx network file to load
verbose	if true is set here some extra debug information will be printed by the runtime

Exceptions

System.IO.FileLoadException	Thrown when the network file could not be loaded
Kiss.EnviromentException	Thrown if the enviroment is unacceptable for the operation of this libaray

References Kiss.Utils.CheckEnvThrow().

Member Function Documentation

Run() [1/2]

AutoResetEvent Kiss.Network.Run ( float[] data, ResultDlg callback )

inline

This method runs an inference pass on the given spectra asynchronously.

Parameters

data	and array of floats containing the data inference on, must be of InputSize length
callback	a delegate that will be called when inference completes, this callback will be called from a native thread in libkissiniferences thread pool

Exceptions

ArgumentException	Thrown when the input is of an incorrect size for the network
InferenceException	Thrown when the inference engine encounteres an internal error

Returns

The AutoResetEvent returned can be waited on to ensure that the inference is finished and the callback has been executed.

Run() [2/2]

AutoResetEvent Kiss.Network.Run ( Spectra spectra, ResultDlg callback )

inline

This method runs an inference pass on the given spectra asynchronously.

Parameters

spectra	the spectra to run inference on, must be of InputSize length
callback	a delegate that will be called when inference completes, this callback will be called from a native thread in libkissiniferences thread pool

Exceptions

ArgumentException	Thrown when the input is of an incorrect size for the network
InferenceException	Thrown when the inference engine encounteres an internal error

Returns

The AutoResetEvent returned can be waited on to ensure that the inference is finished and the callback has been executed.

References Kiss.Spectra.Imaginary, and Kiss.Spectra.Real.

Property Documentation

InputSize

int Kiss.Network.InputSize

get

This Property holds the number of input values the loaded network expects.

For networks with a convolutional input this may be -1 designateing "Any"

OutputMask

bool [] Kiss.Network.OutputMask

getset

This Property is an array of bools enableing or disableing a given output.

This Property must only be set while no inference requests are pending. Disabled outputs will set to zero, exact effect on non-disabled outputs is network depenant. For classifier networks the likelyhoods will reflect the reduction in problem space.