measure.py 文件源码

def evaluate(path):
    system = systems[path]

    measure, scores, clusters_gold, clusters_system = 0., OrderedDict(), [], []

    for lemma in lemmas:
        instances = sorted(gold[lemma].keys())

        senses_gold   = {sid: i for i, sid in enumerate(sorted(set(gold[lemma].values())))}
        senses_system = {sid: i for i, sid in enumerate(sorted(set(system[lemma].values())))}

        clusters_gold   = [senses_gold[gold[lemma][instance]]     for instance in instances]
        clusters_system = [senses_system[system[lemma][instance]] for instance in instances]

        if 'vmeasure' == args.measure:
            if 'instances' == args.average:
                measure += v_measure_score(clusters_gold, clusters_system) * len(instances) / total
            else:
                measure += v_measure_score(clusters_gold, clusters_system)

            scores[lemma] = (
                homogeneity_score(clusters_gold, clusters_system),
                completeness_score(clusters_gold, clusters_system),
                v_measure_score(clusters_gold, clusters_system)
            )
        else:
            scores[lemma] = adjusted_rand_score(clusters_gold, clusters_system)

            if 'instances' == args.average:
                measure += scores[lemma] * len(instances) / total
            else:
                measure += scores[lemma]

    if 'words' == args.average:
        measure /= len(lemmas)

    return measure, scores