ferret.Benchmark.explain#
- Benchmark.explain(text, target=1, show_progress: bool = True, normalize_scores: bool = True, order: int = 1, target_token: str | None = None, target_option: str | None = None) List[Explanation][source]#
Compute explanations using all the explainers stored in the class.
- Parameters:
text (str) – Text string to explain.
target (int) – Class label to produce the explanations for.
show_progress (bool, default False) – Enable progress bar.
normalize_scores (bool, default True) – Apply lp-normalization across tokens to make attribution weights comparable across different explainers.
order (int, default 1) – If normalize_scores=True, this is the normalization order, as passed to numpy.linalg.norm.
- Returns:
List of all explanations produced.
- Return type:
List[Explanation]
Notes
Please reference to User Guide for more information.
Examples
>>> bench = Benchmark(model, tokenizer) >>> explanations = bench.explain("I love your style!", target=2)
Please note that by default we apply L1 normalization across tokens, to make feature attribution weights comparable among explainers. To turn it off, you should use:
>>> bench = Benchmark(model, tokenizer) >>> explanations = bench.explain("I love your style!", target=2, normalize_scores=False)