api/org.diffkt.model/LinearAfterResetGru

LinearAfterResetGru

class LinearAfterResetGru(numInputs: Int, numHidden: Int, initialHidden: DTensor?, accType: RecurrentBase.RecurrentBase.AccType, xh2u: Dense, xh2r: Dense, xh2n: Dense) : GRU

Linear-after-reset GRU

In the computation of the candidate activation vector, the linear transform is applied after the hidden state goes through the reset gate.

\hat{h}t = tanh(W_h x_t + U_h (r_t * h{t-1}) + bias)

Constructors

LinearAfterResetGru

fun LinearAfterResetGru(numInputs: Int, numHidden: Int, random: Random, initialHidden: DTensor? = null, acc: RecurrentBase.RecurrentBase.AccType = AccType.Fold)

LinearAfterResetGru

fun LinearAfterResetGru(numInputs: Int, numHidden: Int, initialHidden: DTensor? = null, accType: RecurrentBase.RecurrentBase.AccType = AccType.Fold, xh2u: Dense, xh2r: Dense, xh2n: Dense)

Functions

accMap

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

cell

open override fun cell(state: Pair<DTensor, DTensor>, x: DTensor): Pair<DTensor, DTensor>

cpu

open override fun cpu(): GRU

doRecurrence

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

Do the recurrence.

equals

open operator override fun equals(other: Any?): Boolean

extractTangent

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

fold

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

hashCode

open override fun hashCode(): Int

invoke

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

load

open override fun load(from: ByteBuffer): GRU

processForBatching

open override fun processForBatching(initialState: DTensor, initialOutput: DTensor, batchSize: Int): Pair<DTensor, 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): GRU

withTrainables

open override fun withTrainables(trainables: List<Trainable<*>>): GRU

wrap

open override fun wrap(wrapper: Wrapper): GRU

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

Properties

accType

open override val accType: RecurrentBase.RecurrentBase.AccType

batchAxis

open val batchAxis: Int

initialHidden

val initialHidden: DTensor? = null

initialOutput

open override val initialOutput: FloatTensor

initialState

open override val initialState: DTensor

numHidden

val numHidden: Int

numInputs

val numInputs: Int

sequenceAxis

open val sequenceAxis: Int

trainables

open override val trainables: List<Trainable<*>>