def test_fmeasure_error(self):
with self.assertRaises(Exception) as context:
metrics.SegmentationFMeasureEvaluator().evaluate_one_sent(
"aaa b".split(), "aaa".split())
t = line.strip()
if post_process is not None:
t = post_process(t)
data.append(t)
return data
eval_shortcuts = {
"bleu": lambda: metrics.BLEUEvaluator(),
"gleu": lambda: metrics.GLEUEvaluator(),
"wer": lambda: metrics.WEREvaluator(),
"cer": lambda: metrics.CEREvaluator(),
"recall": lambda: metrics.RecallEvaluator(),
"accuracy": lambda: metrics.SequenceAccuracyEvaluator(),
"fmeasure": lambda: metrics.FMeasureEvaluator(),
"seg_fmeasure": lambda: metrics.SegmentationFMeasureEvaluator(),
"rnng_parse_fmeasure": lambda: metrics.RNNGParseFMeasureEvaluator(),
}
def xnmt_evaluate(ref_file: Union[str, Sequence[str]],
hyp_file: Union[str, Sequence[str]],
evaluators: Sequence[metrics.Evaluator],
desc: Any = None) -> Sequence[metrics.EvalScore]:
""""Returns the eval score (e.g. BLEU) of the hyp sents using reference trg sents
Args:
ref_file: path of the reference file
hyp_file: path of the hypothesis trg file
evaluators: Evaluation metrics. Can be a list of evaluator objects, or a shortcut string
desc: descriptive string passed on to evaluators
def test_fmeasure(self):
self.assertEqual(
metrics.SegmentationFMeasureEvaluator().evaluate_one_sent(
"ab c def".split(), "a bc def".split()).value(), 0.80)