Struct Trainer 

pub struct Trainer { /* private fields */ }

Main trainer for model training.

Implementations

impl Trainer

pub fn new( config: TrainerConfig, loss_fn: Box<dyn Loss>, optimizer: Box<dyn Optimizer>, ) -> Self

Create a new trainer.

pub fn with_scheduler(self, scheduler: Box<dyn LrScheduler>) -> Self

Set learning rate scheduler.

pub fn with_callbacks(self, callbacks: CallbackList) -> Self

Set callbacks.

pub fn with_metrics(self, metrics: MetricTracker) -> Self

Set metrics.

pub fn train( &mut self, train_data: &ArrayView<'_, f64, Ix2>, train_targets: &ArrayView<'_, f64, Ix2>, val_data: Option<&ArrayView<'_, f64, Ix2>>, val_targets: Option<&ArrayView<'_, f64, Ix2>>, parameters: &mut HashMap<String, Array<f64, Ix2>>, ) -> TrainResult<TrainingHistory>

Train the model.

pub fn get_state(&self) -> &TrainingState

Get current training state.

pub fn save_checkpoint( &self, path: &PathBuf, parameters: &HashMap<String, Array<f64, Ix2>>, history: &TrainingHistory, best_val_loss: Option<f64>, ) -> TrainResult<()>

Save a complete training checkpoint.

This saves all state needed to resume training, including:

• Model parameters
• Optimizer state
• Scheduler state (if present)
• Training history
• Current epoch and losses

pub fn load_checkpoint( &mut self, path: &PathBuf, ) -> TrainResult<(HashMap<String, Array<f64, Ix2>>, TrainingHistory, usize)>

Resume training from a checkpoint.

This restores all training state including parameters, optimizer state, and history. Training will resume from the saved epoch.

Returns the restored parameters, history, and starting epoch.

pub fn train_from_checkpoint( &mut self, checkpoint_path: &PathBuf, train_data: &ArrayView<'_, f64, Ix2>, train_targets: &ArrayView<'_, f64, Ix2>, val_data: Option<&ArrayView<'_, f64, Ix2>>, val_targets: Option<&ArrayView<'_, f64, Ix2>>, ) -> TrainResult<(HashMap<String, Array<f64, Ix2>>, TrainingHistory)>

Train the model starting from a checkpoint.

This is a convenience method that loads a checkpoint and continues training.