Struct DeterministicPhaseTrainer 

pub struct DeterministicPhaseTrainer { /* private fields */ }

Deterministic phase-based trainer.

Orchestrates training with guaranteed deterministic outcomes. Uses warmup → full → predict → correct cycle with residual tracking.

Implementations

impl DeterministicPhaseTrainer

pub fn new( param_shapes: &[(String, Vec<usize>)], config: DeterministicPhaseConfig, device: &Device, ) -> Result<DeterministicPhaseTrainer, OptimError>

Create a new deterministic phase trainer.

Arguments

• param_shapes - List of (name, shape) tuples for parameters
• config - Phase training configuration
• device - Device for tensor storage

Errors

Returns error if predictor initialization fails.

pub fn begin_step(&mut self) -> Result<DeterministicStepInfo, OptimError>

Begin a training step.

Returns information about the current phase and whether backward pass (full gradient computation) is needed.

pub fn needs_backward(&self) -> bool

Check if backward pass is needed for current step.

pub fn record_full_gradients( &mut self, gradients: &HashMap<String, Tensor>, ) -> Result<(), OptimError>

Record full gradients after backward pass.

Called during WARMUP, FULL, or CORRECT phases after computing gradients via backpropagation.

Arguments

• gradients - Map of parameter names to gradient tensors

pub fn get_predicted_gradients( &mut self, ) -> Result<HashMap<String, Tensor>, OptimError>

Get predicted gradients for current step.

Called during PREDICT phase to get deterministic gradient predictions.

Returns

Map of parameter names to predicted gradient tensors.

pub fn end_step(&mut self, loss: f32) -> Result<(), OptimError>

End the current training step.

Updates internal state and statistics.

Arguments

• loss - Loss value for this step

pub const fn current_phase(&self) -> DeterministicPhase

Get current training phase.

pub const fn warmup_complete(&self) -> bool

Check if warmup is complete.

pub fn get_stats(&self) -> DeterministicTrainerStats

Get training statistics.

pub fn reset(&mut self) -> Result<(), OptimError>

Reset trainer state.