KGraph

class KGraph(graphDef: ByteArray, prefix: String) : AutoCloseable

A wrapper object to extend functionality and behaviour of static TensorFlow graph.

It tracks all model variables (used in optimizers or layers) and its initializers.

Parameters

graphDef

A serialized representation of the graph.

prefix

A prefix that will be prepended to names in graphDef.

Constructors

KGraph

fun KGraph(graphDef: ByteArray)

KGraph

fun KGraph(graphDef: ByteArray, prefix: String)

Creates KGraph by serialized representation of the graph.

Functions

addInitializer

fun addInitializer(variableName: String, initializer: Assign<Float>)

Adds an initializer for layer variable tracked in KGraph.

addLayerVariable

fun addLayerVariable(variable: Variable<Float>, isTrainable: Boolean)

Adds a variable used in layer to the pool of tracked variables.

addOptimizerVariable

fun addOptimizerVariable(variable: Variable<Float>)

Adds a variable used in optimizer to the pool of tracked variables.

addOptimizerVariableAssignAddInitializer

fun addOptimizerVariableAssignAddInitializer(initializer: AssignAdd<Float>)

Adds an optimizer initializer of special 'AssignAdd' type for optimizer variable tracked in KGraph.

addOptimizerVariableInitializer

fun addOptimizerVariableInitializer(initializer: Assign<*>)

Adds an optimizer initializer for optimizer variable tracked in KGraph.

addVariableRegularizer

fun addVariableRegularizer(variable: Variable<Float>, regularizer: Regularizer)

Adds a variable related to the given regularizer to the pool of tracked variables.

open override fun close()

Closes internal TensorFlow graph.

copy

fun copy(): KGraph

Makes a graph copy.

frozenLayerVariables

fun frozenLayerVariables(): List<Variable<Float>>

Returns a list of non-trainable, 'frozen' variables used in layers.

initializeGraphVariables

fun initializeGraphVariables(session: Session)

Initializes TensorFlow graph variables used in model layers.

initializeOptimizerVariables

fun initializeOptimizerVariables(session: Session)

Initializes TensorFlow graph variables used in optimizer.

isVariableTrainable

fun isVariableTrainable(variableName: String): Boolean

Returns the variable's ability to be changed during the training.

layerVariables

fun layerVariables(): List<Variable<Float>>

Returns all variables used in all model layers.

optimizerVariables

fun optimizerVariables(): List<Variable<Float>>

Returns all variables used in optimizer and initialized by Assign TensorFlow operand.

toString

open override fun toString(): String

trainableLayerVariables

fun trainableLayerVariables(): List<Variable<Float>>

Returns a list of trainable variables used in layers.

variableNames

fun variableNames(): List<String>

Returns list of variable names in TensorFlow graph.

Properties

isClosed

var isClosed: Boolean = false

True if the graph object is closed and the occupied resources are freed.