oa::RnnCell

/ Rnn — vanilla Elman recurrent module. h_new = tanh( W_ih x + b_ih + W_hh h + b_hh ) Mirrors the Gru / GruCell API 1:1 (zeroState / step / forward, stacked layers) so the two recurrent modules are interchangeable in a model. Like the GRU it fuses its pointwise tail (Add + Tanh) into one RnnCellPointwise kernel. The two Linear projections stay separate dispatches; since oa::FnMatrix::Linear is pure dispatch, each needs a manual grad-node attach — see AttachLinearGrad in Rnn.cpp.

ClC++ public headers

Inheritance

public Module

Public Methods

oa::Matrix oa::RnnCell::forward(const oa::Matrix & inInput)

oa::I32 oa::RnnCell::hiddenSize()

oa::Matrix oa::RnnCell::inputProjection(const oa::Matrix & inInput)

oa::I32 oa::RnnCell::inputSize()

oa::Matrix oa::RnnCell::step(const oa::Matrix & inInput, const oa::Matrix & inHidden)

oa::Matrix oa::RnnCell::stepWithGi(const oa::Matrix & inGi, const oa::Matrix & inHidden, oa::U32 inTimeOffset = 0, oa::U32 inBatchStride = 1)

oa::Matrix oa::RnnCell::zeroState(oa::I32 inBatch)

Constructor & Destructor Documentation

oa::RnnCell::RnnCell(
	oa::I32 inInputSize,
	oa::I32 inHiddenSize,
	bool inBias = true
)

No public source comment is attached to this declaration.

Parameters

inInputSize

oa::I32

—

inHiddenSize

oa::I32

—

inBias

bool

Default: true

Public Method Documentation

oa::Matrix oa::RnnCell::forward(
	const oa::Matrix & inInput
)

No public source comment is attached to this declaration.

Parameters

inInput

const oa::Matrix &

—

Returns

oa::Matrix

The declared return value.

oa::I32 oa::RnnCell::hiddenSize()

No public source comment is attached to this declaration.

Returns

oa::I32

The declared return value.

oa::Matrix oa::RnnCell::inputProjection(
	const oa::Matrix & inInput
)

split projections so Rnn can hoist the (recurrence-free) input projection out of the timestep loop into one batched GEMM. inputProjection: gi = Linear(x, W_ih, b_ih) for any row count → [*, H]. stepWithGi: consumes a precomputed gi [B*T, H] at row offset timeOffset and runs only the recurrent gh = Linear(h, W_hh) + fused tanh pointwise (no per-step Slice).

Parameters

inInput

const oa::Matrix &

—

Returns

oa::Matrix

The declared return value.

oa::I32 oa::RnnCell::inputSize()

No public source comment is attached to this declaration.

Returns

oa::I32

The declared return value.

oa::Matrix oa::RnnCell::step(
	const oa::Matrix & inInput,
	const oa::Matrix & inHidden
)

No public source comment is attached to this declaration.

Parameters

inInput

const oa::Matrix &

—

inHidden

const oa::Matrix &

—

Returns

oa::Matrix

The declared return value.

oa::Matrix oa::RnnCell::stepWithGi(
	const oa::Matrix & inGi,
	const oa::Matrix & inHidden,
	oa::U32 inTimeOffset = 0,
	oa::U32 inBatchStride = 1
)

No public source comment is attached to this declaration.

Parameters

inGi

const oa::Matrix &

—

inHidden

const oa::Matrix &

—

inTimeOffset

oa::U32

Default: 0

inBatchStride

oa::U32

Default: 1

Returns

oa::Matrix

The declared return value.

oa::Matrix oa::RnnCell::zeroState(
	oa::I32 inBatch
)

No public source comment is attached to this declaration.

Parameters

inBatch

oa::I32

—

Returns

oa::Matrix

The declared return value.