Global Methods

Global counterfactual methods find universal transformations that apply across an entire dataset or population. They answer: "What systematic changes would alter predictions for many instances?"

Available Methods

Method	Description	Key Feature
GLOBE-CE	Global counterfactual explanations	Dataset-wide transformations
AReS	Anchor/rule-based explanations	Interpretable rules

When to Use Global Methods

Global methods are ideal when you need to:

• Understand systematic model behavior
• Identify policy-level interventions
• Find transformations that work for many instances
• Gain high-level insights into the model

Comparison with Local Methods

Aspect	Local Methods	Global Methods
Scope	Single instance	Entire dataset
Output	Individual counterfactual	Universal transformation
Use case	Personal recourse	Policy insights
Interpretability	Instance-specific	Broadly applicable

Example Usage

from counterfactuals.cf_methods.global_methods import GLOBECE

# Initialize method
method = GLOBECE(
    gen_model=flow_model,
    disc_model=classifier,
    disc_model_criterion=criterion,
    device="cuda"
)

# Find global counterfactual transformation
result = method.explain(
    X=X_test,
    y_origin=y_test,
    y_target=target_class,
    X_train=X_train,
    y_train=y_train
)

# The transformation applies to multiple instances
print(f"Global transformation found for {len(X_test)} instances")