SToG.trainer

Training utilities for feature selection.

Classes

FeatureSelectionTrainer(model, selector, ...)

Trainer with proper lambda search and early stopping.

class SToG.trainer.FeatureSelectionTrainer(model, selector, criterion, lambda_reg=0.1, device='cpu')

Bases: object

Trainer with proper lambda search and early stopping. Handles joint training of classification model and feature selector.

__init__(model, selector, criterion, lambda_reg=0.1, device='cpu')

Initialize trainer.

Parameters:

• model – Classification model (nn.Module)

• selector – Feature selector (BaseFeatureSelector)

• criterion – Loss function

• lambda_reg – Regularization strength

• device – Device to run on

train_epoch(X_train, y_train, X_val, y_val)

Train for one epoch.

fit(X_train, y_train, X_val, y_val, epochs=300, patience=50, verbose=False)

Train the model with early stopping.

Parameters:

• X_train – Training features [N_train, D]

• y_train – Training labels [N_train]

• X_val – Validation features [N_val, D]

• y_val – Validation labels [N_val]

• epochs – Maximum number of epochs

• patience – Early stopping patience

• verbose – Whether to print progress

Returns:

Training history dictionary

evaluate(X_test, y_test)

Evaluate on test set.

Parameters:

• X_test – Test features [N_test, D]

• y_test – Test labels [N_test]

Returns:

Dictionary with test metrics