ferret.TauLOO_Evaluation#
- class ferret.TauLOO_Evaluation(model, tokenizer, task_name)[source]#
- __init__(model, tokenizer, task_name)#
Methods
__init__(model, tokenizer, task_name)compute_evaluation(explanation, ...)Evaluate an explanation on the tau-LOO metric, i.e., the Kendall tau correlation between the explanation scores and leave one out (LOO) scores, computed by leaving one feature out and computing the change in the prediciton probability
Attributes
BEST_VALUELOWER_IS_BETTERMAX_VALUEMETRIC_FAMILYMIN_VALUENAMESHORT_NAMEtokenizer