RecurrentBase

interface RecurrentBase<Recurrent : RecurrentBase<Recurrent, T>, T> : TrainableLayer<Recurrent>

Types

RecurrentBase

object RecurrentBase

Functions

accMap

open fun accMap(t: DTensor, sequenceAxis: Int, initialState: T): Pair<T, DTensor>

cell

abstract fun cell(state: Pair<T, DTensor>, x: DTensor): Pair<T, DTensor>

cpu

open override fun cpu(): Recurrent

doRecurrence

open fun doRecurrence(x: DTensor, initialState: T = this.initialState): DTensor

Do the recurrence.

extractTangent

open override fun extractTangent(output: DTensor, extractor: (DTensor, DTensor) -> DTensor): TrainableComponent.Companion.Tangent

fold

open fun fold(t: DTensor, sequenceAxis: Int, initialState: T): Pair<T, DTensor>

getSingleInput

open fun getSingleInput(inputs: Array<out DTensor>): DTensor

Helper to check that the layer was called with a single input. Returns that input if successful, else errors.

gpu

open override fun gpu(): Recurrent

invoke

open operator override fun invoke(vararg inputs: DTensor): DTensor

load

open override fun load(from: ByteBuffer): Recurrent

processForBatching

abstract fun processForBatching(initialState: T, initialOutput: DTensor, batchSize: Int): Pair<T, DTensor>

store

open override fun store(into: ByteBuffer): ByteBuffer

open fun to(device: Device): OnDevice

trainingStep

open override fun trainingStep(optim: Optimizer<*>, tangent: Trainable.Tangent): Recurrent

withTrainables

abstract fun withTrainables(trainables: List<Trainable<*>>): Recurrent

wrap

open override fun wrap(wrapper: Wrapper): Recurrent

The wrap function should return the same static type it is declared on.

Properties

accType

abstract val accType: RecurrentBase.RecurrentBase.AccType

batchAxis

open val batchAxis: Int

initialOutput

abstract val initialOutput: DTensor

initialState

abstract val initialState: T

sequenceAxis

open val sequenceAxis: Int

trainables

abstract val trainables: List<Trainable<*>>

Inheritors

GRU