oa::ItRolloutTraining

Coordinates the two-phase lifecycle of synchronous on-policy training. Environment stepping, policy evaluation, and loss construction remain caller supplied; an owned `ItTraining` controls each exact optimizer update.

ClC++ public headers

Public Methods

oa::Status oa::ItRolloutTraining::abortRollout(RolloutBuffer & inRollout)

oa::Status oa::ItRolloutTraining::beginRollout(RolloutBuffer & inRollout)

bool oa::ItRolloutTraining::beginUpdate()

const ItRolloutTrainingConfig & oa::ItRolloutTraining::config()

oa::Status oa::ItRolloutTraining::finalizeRollout(RolloutBuffer & inRollout, const GaeConfig & inConfig = {})

oa::Status oa::ItRolloutTraining::finish()

bool oa::ItRolloutTraining::isDone()

bool oa::ItRolloutTraining::isValid()

const oa::Status & oa::ItRolloutTraining::lastStatus()

oa::Status oa::ItRolloutTraining::nextUpdate(const oa::Matrix & inLoss)

RolloutTrainingPhase oa::ItRolloutTraining::phase()

oa::U32 oa::ItRolloutTraining::rolloutIndex()

oa::U32 oa::ItRolloutTraining::updateEpoch()

const oa::ItTraining & oa::ItRolloutTraining::updateLoop()

oa::ItTraining & oa::ItRolloutTraining::updateLoop()

Constructor & Destructor Documentation

oa::ItRolloutTraining::ItRolloutTraining(
	oa::Engine & inEngine,
	oa::Optimizer & inOptimizer,
	const ItRolloutTrainingConfig & inConfig
)

No public source comment is attached to this declaration.

Parameters

inEngine

oa::Engine &

—

inOptimizer

oa::Optimizer &

—

inConfig

const ItRolloutTrainingConfig &

—

Public Method Documentation

oa::Status oa::ItRolloutTraining::abortRollout(
	RolloutBuffer & inRollout
)

Restores the pre-beginRollout collect state after the caller cancels the unsubmitted command transaction. This is also valid after finalizeRollout and before the first update begins.

Parameters

inRollout

RolloutBuffer &

—

Returns

oa::Status

The declared return value.

oa::Status oa::ItRolloutTraining::beginRollout(
	RolloutBuffer & inRollout
)

Starts one collection cycle and resets the supplied fixed-capacity buffer.

Parameters

inRollout

RolloutBuffer &

—

Returns

oa::Status

The declared return value.

bool oa::ItRolloutTraining::beginUpdate()

Must be called immediately before recording one differentiable PPO update. It advances the underlying ItTraining lifecycle and returns false only on invalid phase/configuration or after completion.

Returns

bool

The declared return value.

const ItRolloutTrainingConfig & oa::ItRolloutTraining::config()

No public source comment is attached to this declaration.

Returns

const ItRolloutTrainingConfig &

The declared return value.

oa::Status oa::ItRolloutTraining::finalizeRollout(
	RolloutBuffer & inRollout,
	const GaeConfig & inConfig = {}
)

Requires a full buffer, records GAE, and opens the update phase.

Parameters

inRollout

RolloutBuffer &

—

inConfig

const GaeConfig &

Default: {}

Returns

oa::Status

The declared return value.

oa::Status oa::ItRolloutTraining::finish()

No public source comment is attached to this declaration.

Returns

oa::Status

The declared return value.

bool oa::ItRolloutTraining::isDone()

No public source comment is attached to this declaration.

Returns

bool

The declared return value.

bool oa::ItRolloutTraining::isValid()

No public source comment is attached to this declaration.

Returns

bool

The declared return value.

const oa::Status & oa::ItRolloutTraining::lastStatus()

No public source comment is attached to this declaration.

Returns

const oa::Status &

The declared return value.

oa::Status oa::ItRolloutTraining::nextUpdate(
	const oa::Matrix & inLoss
)

Completes the update recorded after beginUpdate: optimizer, submit, sync, metrics and phase advancement.

Parameters

inLoss

const oa::Matrix &

—

Returns

oa::Status

The declared return value.

RolloutTrainingPhase oa::ItRolloutTraining::phase()

No public source comment is attached to this declaration.

Returns

RolloutTrainingPhase

The declared return value.

oa::U32 oa::ItRolloutTraining::rolloutIndex()

No public source comment is attached to this declaration.

Returns

oa::U32

The declared return value.

oa::U32 oa::ItRolloutTraining::updateEpoch()

No public source comment is attached to this declaration.

Returns

oa::U32

The declared return value.

const oa::ItTraining & oa::ItRolloutTraining::updateLoop()

No public source comment is attached to this declaration.

Returns

const oa::ItTraining &

The declared return value.

oa::ItTraining & oa::ItRolloutTraining::updateLoop()

No public source comment is attached to this declaration.

Returns

oa::ItTraining &

The declared return value.