Python手动计算BLEU值（可运行）

Python手动计算BLEU（可运行）

话不多说，直接上代码

import math

def count_ngram(candidate, references, n):
    clipped_count = 0
    count = 0
    r = 0
    c = 0
    for si in range(len(candidate)):
        # 计算每个句子的精度
        ref_counts = []
        ref_lengths = []
        # 构建 ngram 计数字典
        for reference in references:
            ref_sentence = reference[si]
            ngram_d = {}
            words = ref_sentence.strip().split()
            ref_lengths.append(len(words))
            limits = len(words) - n + 1
            # 遍历句子考虑 ngram 长度
            for i in range(limits):
                ngram = ' '.join(words[i:i+n]).lower()
                if ngram in ngram_d.keys():
                    ngram_d[ngram] += 1
                else:
                    ngram_d[ngram] = 1
            ref_counts.append(ngram_d)
        # 候选句
        cand_sentence = candidate[si]
        cand_dict = {}
        words = cand_sentence.strip().split()
        limits = len(words) - n + 1
        for i in range(0, limits):
            ngram = ' '.join(words[i:i + n]).lower()
            if ngram in cand_dict:
                cand_dict[ngram] += 1
            else:
                cand_dict[ngram] = 1
        clipped_count += clip_count(cand_dict, ref_counts)
        count += limits
        r += best_length_match(ref_lengths, len(words))
        c += len(words)
    if clipped_count == 0:
        pr = 0
    else:
        pr = float(clipped_count) / count
    bp = brevity_penalty(c, r)
    return pr, bp

def clip_count(cand_d, ref_ds):
    """考虑所有引用，计算每个 ngram 的剪辑计数"""
    count = 0
    for m in cand_d.keys():
        m_w = cand_d[m]
        m_max = 0
        for ref in ref_ds:
            if m in ref:
                m_max = max(m_max, ref[m])
        m_w = min(m_w, m_max)
        count += m_w
    return count

def best_length_match(ref_l, cand_l):
    """找到最接近候选句的参考长度"""
    least_diff = abs(cand_l-ref_l[0])
    best = ref_l[0]
    for ref in ref_l:
        if abs(cand_l-ref) < least_diff:
            least_diff = abs(cand_l-ref)
            best = ref
    return best

def brevity_penalty(c, r):
    if c > r:
        bp = 1
    else:
        bp = math.exp(1-(float(r)/c))
    return bp

def BLEU(candidate, references, n_gram):
    pr, bp = count_ngram(candidate, references, n_gram)
    return pr

candidate = ['我', '爱', '你']

references = [['我', '爱', '你']]

bleu = BLEU(candidate, references, 1)
bleu

其中1代表n-gram，运行结果：