A graph is a collection of buffers and nodes that together form a neural network. The graph is created from a json definition file. After creation, the contents of the graph can be loaded from or saved to disk. More...

Inheritance diagram for SmartEngine.NeuralNetworks.Graph:

Classes
class	CInfo
	Constructor info More...

Public Member Functions
	Graph (CInfo cinfo)

Context	GetContext ()
	Returns the context associated with the graph More...

Graph	DeepCopy ()
	Copies the internal structure of the graph into a new graph More...

Graph	Instance ()
	Creates a copy of this graph with all internal weights referenced to this graph. More...

unsafe void	BindComponentInputs (ComponentInputBinding[] bindings)
	Binds the component inputs in the graph to the specified set of nodes. Any component inputs not included will be bound to buffer inputs. More...

void	CopyWeightsFrom (Graph graph, float percent=1.0f)
	Copies the neuron layer weights from the specified graph into this graph. The specified graph must have the same exact structure as this graph. More...

void	SetRandomWeights ()
	Initialize the weights of all trainable layers to random values. More...

T	GetNode< T > (string name)
	Returns the specified node as the specified type. Null is returned if the node is not found or is of a different type. More...

GraphNode	GetOutputNode (int index)
	Returns an output node of the graph as a GraphNode. The nodes are ordered in the appearance in the graph definition or the order they were added manually. More...

void	AddNode (GraphNode node)
	Adds a node to the graph. This will be considered an output node until a node connecting to this one is added to the graph. More...

float	GetWeightStandardDeviation ()
	Returns the average standard deviation of all trainable weights in the graph. More...

void	ResetNeuronLayers ()
	Calls Reset() on all trainable layers in the graph. This is only necessary for LSTM neuron layers. More...

void	StepNeuronLayers (bool autoReset=true)
	Calls Step() on all trainable layers in the graph. This is only necessary for LSTM neuron layers. More...

Public Member Functions inherited from SmartEngine.Resource
SerializationResult	Load (string appendName=null)
	Load this object from disk. More...

SerializationResult	Save (string appendName=null)
	Save this object to disk. More...

SerializationResult	Serialize (MemoryBuffer buffer)
	Write the contents of this object to a buffer. More...

SerializationResult	Serialize (out byte[] data)
	Write the contents of this object to a byte array. More...

SerializationResult	Deserialize (MemoryBuffer buffer)
	Fill this object with contents from a buffer. More...

SerializationResult	Deserialize (byte[] data)
	Fill this object with contents from a byte array. More...

Public Member Functions inherited from SmartEngine.Object
void	AddRef ()
	Increments the internal reference count on this object. It is not common to use this method directly. More...

void	Release ()
	Decrements the internal reference count on this object. It is not common to use this method directly. More...

Public Member Functions inherited from SmartEngine.Disposable
void	Dispose ()
	Cleans up any internal state. It is not safe to use an object after it has been disposed. More...

Static Public Member Functions
static Graph	CreateFromDefinitionString (CInfo cinfo, string graphDefinition)
	Creates a new graph from a graph definition json string More...

static Graph	CreateFromResource (CInfo cinfo, string resourceName=null)
	Creates and loads a new Graph from a resource. More...

Properties
string	Name `[get]`
	Returns the name of the graph. More...

int	SequenceLength `[get, set]`
	Gets or sets the desired sequence length for this graph. This is after how many steps the graph should be reset. Only applicable to graphs that need to be stepped (such as LSTMs). More...

Properties inherited from SmartEngine.Resource
string	ResourceName `[get]`
	Returns the name of this resource passed to the constructor. More...

SerializationResult	LastLoadResult `[get]`
	Returns the result of the last call to Load(). Useful for checking loaded data state after creation. More...

Detailed Description

A graph is a collection of buffers and nodes that together form a neural network. The graph is created from a json definition file. After creation, the contents of the graph can be loaded from or saved to disk.

When loading from a resource, two assets will try to load. One will be the resourceName + "Definition", which should be a json text asset. The other will be resourceName + "Data", which will be the graph data. The definition is required, but the data can be missing on creation.

Member Function Documentation

◆ AddNode()

void SmartEngine.NeuralNetworks.Graph.AddNode ( GraphNode node )

Adds a node to the graph. This will be considered an output node until a node connecting to this one is added to the graph.

Parameters

node

◆ BindComponentInputs()

unsafe void SmartEngine.NeuralNetworks.Graph.BindComponentInputs ( ComponentInputBinding[] bindings )

Binds the component inputs in the graph to the specified set of nodes. Any component inputs not included will be bound to buffer inputs.

This should only be called once before first evaluation. It is not necessary to call this if the graph is used as a graph component or added as a standard graph model to a graph manager.

Called automatically by IGraphComponent upon creation.

Any created buffer inputs will have the name "[Comp Input Name]Buffer" and will be added to the graph. (Ex: "MyCompInput" would have a buffer "MyCompInputBuffer" that can be queried through normal means)

Parameters

bindings
bindingCount

◆ CopyWeightsFrom()

void SmartEngine.NeuralNetworks.Graph.CopyWeightsFrom	(	Graph	graph,
		float	percent = `1.0f`
	)

Copies the neuron layer weights from the specified graph into this graph. The specified graph must have the same exact structure as this graph.

Parameters

graph

◆ CreateFromDefinitionString()

static Graph SmartEngine.NeuralNetworks.Graph.CreateFromDefinitionString	(	CInfo	cinfo,
		string	graphDefinition
	)

static

Creates a new graph from a graph definition json string

Parameters

cinfo
graphDefinition

Returns

◆ CreateFromResource()

static Graph SmartEngine.NeuralNetworks.Graph.CreateFromResource	(	CInfo	cinfo,
		string	resourceName = `null`
	)

static

Creates and loads a new Graph from a resource.

Parameters

cinfo
folder
name

Returns

◆ DeepCopy()

Graph SmartEngine.NeuralNetworks.Graph.DeepCopy ( )

Copies the internal structure of the graph into a new graph

Returns

◆ GetContext()

Context SmartEngine.NeuralNetworks.Graph.GetContext ( )

Returns the context associated with the graph

Returns

◆ GetNode< T >()

T SmartEngine.NeuralNetworks.Graph.GetNode< T > ( string name )

Returns the specified node as the specified type. Null is returned if the node is not found or is of a different type.

Template Parameters

T

Parameters

name

Returns

Type Constraints

T

:

GraphNode

◆ GetOutputNode()

GraphNode SmartEngine.NeuralNetworks.Graph.GetOutputNode ( int index )

Returns an output node of the graph as a GraphNode. The nodes are ordered in the appearance in the graph definition or the order they were added manually.

Parameters

index

Returns

◆ GetWeightStandardDeviation()

float SmartEngine.NeuralNetworks.Graph.GetWeightStandardDeviation ( )

Returns the average standard deviation of all trainable weights in the graph.

Returns

◆ Instance()

Graph SmartEngine.NeuralNetworks.Graph.Instance ( )

Creates a copy of this graph with all internal weights referenced to this graph.

◆ ResetNeuronLayers()

void SmartEngine.NeuralNetworks.Graph.ResetNeuronLayers ( )

Calls Reset() on all trainable layers in the graph. This is only necessary for LSTM neuron layers.

◆ SetRandomWeights()

void SmartEngine.NeuralNetworks.Graph.SetRandomWeights ( )

Initialize the weights of all trainable layers to random values.

◆ StepNeuronLayers()

void SmartEngine.NeuralNetworks.Graph.StepNeuronLayers ( bool autoReset = true )

Calls Step() on all trainable layers in the graph. This is only necessary for LSTM neuron layers.

Parameters

autoReset If true, ResetNeuronLayers() will automatically be called periodically based on the sequence length.

Property Documentation

◆ Name

string SmartEngine.NeuralNetworks.Graph.Name

get

Returns the name of the graph.

◆ SequenceLength

int SmartEngine.NeuralNetworks.Graph.SequenceLength

getset

Gets or sets the desired sequence length for this graph. This is after how many steps the graph should be reset. Only applicable to graphs that need to be stepped (such as LSTMs).

Classes

Public Member Functions

Static Public Member Functions

Properties

Detailed Description

Member Function Documentation

◆ AddNode()

◆ BindComponentInputs()

◆ CopyWeightsFrom()

◆ CreateFromDefinitionString()

◆ CreateFromResource()

◆ DeepCopy()

◆ GetContext()

◆ GetNode< T >()

◆ GetOutputNode()

◆ GetWeightStandardDeviation()

◆ Instance()

◆ ResetNeuronLayers()

◆ SetRandomWeights()

◆ StepNeuronLayers()

Property Documentation

◆ Name

◆ SequenceLength