pos_eval.py 文件源码

def score_icon_plain(ref_file, hyp_file, n_significance_tests=20):

    ref_tags = read_tag_file(ref_file)
    hyp_tags = read_tag_file(hyp_file)

    assert len(ref_tags) == len(hyp_tags), 'ref file and hyp file must have the same number of tags'
    for ref_line, hyp_line in zip(ref_tags, hyp_tags):
        assert len(ref_line) == len(hyp_line), 'ref line and hyp line must have the same number of tags'

    # flatten out tags
    flat_ref_tags = [t for l in ref_tags for t in l]
    flat_hyp_tags = [t for l in hyp_tags for t in l]

    actual_class_f1 = f1_score(flat_ref_tags, flat_hyp_tags, average=None)
    actual_average_f1 = weighted_fmeasure(flat_ref_tags, flat_hyp_tags)

    # END EVALUATION
    return [actual_class_f1, actual_average_f1]