Struct StreamingOptimizer 

pub struct StreamingOptimizer<O, A, D>
where
    A: Float + Send + Sync + ScalarOperand + Debug,
    D: Dimension,
    O: Optimizer<A, D>,
{
    pub data_buffer: VecDeque<StreamingDataPoint<A>>,
    pub step_count: usize,
    /* private fields */
}

Streaming gradient descent optimizer

Fields

data_buffer: VecDeque<StreamingDataPoint<A>>

Data buffer for mini-batches

step_count: usize

Current step count

Implementations

impl<O, A, D> StreamingOptimizer<O, A, D>

where A: Float + Default + Clone + Send + Sync + Debug + ScalarOperand + Sum + DivAssign, D: Dimension, O: Optimizer<A, D> + Send + Sync,

pub fn new(baseoptimizer: O, config: StreamingConfig) -> Result<Self>

Create a new streaming optimizer

pub fn process_sample( &mut self, data_point: StreamingDataPoint<A>, ) -> Result<Option<Array1<A>>>

Process a single streaming data point

pub fn set_lr_adaptation_interval(&mut self, interval: Duration)

Throttle learning-rate adaptation to at most once per interval.

Duration::ZERO (the default) adapts on every processed batch.

pub fn lr_adaptation_interval(&self) -> Duration

The configured learning-rate adaptation interval.

pub fn get_metrics(&self) -> &StreamingMetrics

Get current streaming metrics

pub fn last_gradient(&self) -> Option<&Array1<A>>

The gradient most recently applied to the parameters, if any.

pub fn current_parameters(&self) -> Option<&Array1<A>>

Get the live model parameters currently maintained by the streaming optimizer, if any samples have been processed yet (F27: parameters are tracked on the optimizer itself, not recomputed from nothing).

pub fn is_healthy(&self) -> StreamingHealthStatus

Check if streaming optimizer is healthy (within budgets)

pub fn flush(&mut self) -> Result<Option<Array1<A>>>

Force processing of current buffer

impl<O, A, D> StreamingOptimizer<O, A, D>

where A: Float + Default + Clone + Send + Sync + Debug + ScalarOperand + Sum + DivAssign, D: Dimension, O: Optimizer<A, D> + Send + Sync,

pub fn stream_synchronization_skew_ms(&self) -> Option<f64>

Measured inter-stream synchronization skew (ms), if multi-stream coordination is enabled and at least two streams have been seen.

pub fn registered_stream_count(&self) -> usize

Number of distinct logical streams observed so far.

pub fn multi_stream_coordinator(&self) -> Option<&MultiStreamCoordinator<A>>

Read-only access to the multi-stream coordinator.

pub fn add_pipeline_stage( &mut self, stage_id: impl Into<String>, processing_function: &str, ) -> Result<()>

Append a processing stage to the pipeline every batch flows through. The stage function name is validated immediately.

pub fn pipeline(&self) -> &PipelineExecutionManager<A>

Read-only access to the pipeline manager and its measured stage metrics.

pub fn fusion_optimizer_mut(&mut self) -> Option<&mut StreamFusionOptimizer<A>>

Mutable access to the stream fusion optimizer, for selecting a fusion strategy or consensus algorithm.

pub fn fusion_optimizer(&self) -> Option<&StreamFusionOptimizer<A>>

Read-only access to the stream fusion optimizer.

pub fn predictive_engine(&self) -> Option<&PredictiveStreamingEngine<A>>

Read-only access to the predictive engine.

pub fn prediction_confidence(&self) -> A

Measured confidence of the predictive engine in [0, 1]; zero while the engine is disabled or untrained.

pub fn normalized_prediction_error(&self) -> A

Measured normalized one-step prediction error of the predictive engine.

pub fn forecast_len(&self) -> usize

Number of points in the predictive engine’s most recent forecast.

pub fn qos_status(&self) -> &QoSStatus

The most recent QoS evaluation.

pub fn qos_pressure(&self) -> f64

How far outside its service level objectives the system currently is.

pub fn resource_allocation(&self) -> Option<&ResourceAllocation>

The most recent resource allocation decision, if adaptive resource allocation is enabled.

pub fn realtime_result(&self) -> Option<&RTOptimizationResult>

The most recent real-time optimization outcome.

pub fn realtime_optimizer(&self) -> &RealTimeOptimizer

Read-only access to the real-time optimizer and its measurements.

pub fn cpu_duty_cycle_percent(&self) -> f64

Measured fraction of wall-clock time this optimizer has spent actually processing batches, as a percentage. A real duty-cycle measurement, not a guess at system-wide CPU load.