Training pipeline

fyt.core.training_pipeline

Training pipeline orchestrator.

Sequences all pipeline stages using protocol-based dependencies, enabling component substitution without modifying this module.

TrainingPipeline

Complete training pipeline orchestrating all components.

This class integrates data management, preprocessing, feature selection, and model training into a unified pipeline with dependency injection.

All component dependencies are declared as protocols, allowing any conforming implementation to be substituted.

experiment_reporter property

The experiment reporter instance.

model property

The trained model from the trainer.

Returns:

Type	Description
	The trained model instance.

__init__(config, data_manager, pre_processor, target_processor, feature_selector, trainer, metrics_evaluator, hyperparameter_optimizer=None, experiment_reporter=None, task=TaskType.CLASSIFICATION)

Initialize the TrainingPipeline.

Parameters:

Name	Type	Description	Default
config	TrainerPipelineConfig	Configuration for the entire training pipeline.	required
data_manager	DataSplitter	Data loader and splitter.	required
pre_processor	Preprocessor	Feature preprocessor.	required
target_processor	TargetTransformer	Target variable encoder.	required
feature_selector	FeatureFilter	Feature selection strategy.	required
trainer	Predictor	Model trainer and predictor.	required
metrics_evaluator	MetricsEvaluator	Metrics computation.	required
hyperparameter_optimizer	HyperparameterOptimizer | None	Optional hyperparameter optimizer.	None
experiment_reporter	ExperimentReporter | None	Optional experiment reporter for logging results.	None
task	TaskType	Learning task type.	CLASSIFICATION

get_class_mapping()

Get the class mapping from the target processor.

Returns:

Type	Description
dict[str, int] | None	Dictionary mapping class labels to integers or None if not fitted yet.

get_selected_features()

Get the list of features selected by the feature selector.

Returns:

Type	Description
list[str] | None	List of selected feature names or None if not fitted yet.

run(test_size=0.2, random_state=42)

Run the complete training pipeline.

Parameters:

Name	Type	Description	Default
test_size	float	Proportion of data to use for testing.	0.2
random_state	int	Random seed for reproducibility.	42

Returns:

Type	Description
MetricResults	MetricResults containing all evaluation metrics.

to_inference_pipeline()

Bundle the fitted components into a persistable InferencePipeline.

Returns:

Type	Description
	An InferencePipeline wrapping the fitted preprocessor, feature
	selector, target processor, and trained model.