Struct TrainingSimulation 

pub struct TrainingSimulation { /* private fields */ }

Simulated training scenario for reproducible ML experiments.

Implements Toyota Way principles:

• Jidoka: Stop-on-anomaly via AnomalyDetector
• Heijunka: Load-balanced batch iteration
• Kaizen: Continuous improvement tracking

Implementations

impl TrainingSimulation

pub fn new(seed: u64) -> Self

Create new training simulation with deterministic seed.

pub fn with_config(seed: u64, config: TrainingConfig) -> Self

Create with custom configuration.

pub fn set_anomaly_detector(&mut self, detector: AnomalyDetector)

Set anomaly detector.

pub fn config(&self) -> &TrainingConfig

Get current training configuration.

pub fn trajectory(&self) -> &TrainingTrajectory

Get current trajectory.

pub fn step( &mut self, loss: f64, gradient_norm: f64, ) -> SimResult<Option<TrainingState>>

Simulate a single training step with given loss and gradient norm.

This is a simplified simulation - real training would compute actual forward/backward passes. This enables testing training dynamics without actual model computation.

Errors

Returns error if a training anomaly is detected (Jidoka).

pub fn simulate<F>( &mut self, epochs: u64, loss_fn: F, ) -> SimResult<&TrainingTrajectory>

where F: FnMut(u64, &mut SimRng) -> (f64, f64),

Simulate training for specified epochs using a loss function.

The loss_fn takes (epoch, rng) and returns (loss, gradient_norm).

Errors

Returns error if a training anomaly is detected.

pub fn replay_from(&mut self, checkpoint: &TrainingState) -> SimResult<()>

Replay training from a checkpoint state.

Errors

Returns error if RNG state restoration fails.

pub fn journal(&self) -> &EventJournal

Get the event journal.

pub fn reset(&mut self, seed: u64)

Reset simulation state.