BLEU评分在Python中的实现

Question

class BLEU(object):
    def compute(candidate, references, weights):
        candidate = [c.lower() for c in candidate]
        references = [[r.lower() for r in reference] for reference in references]

        p_ns = (BLEU.modified_precision(candidate, references, i) for i, _ in enumerate(weights, start=1))
        s = math.fsum(w * math.log(p_n) for w, p_n in zip(weights, p_ns) if p_n)

        bp = BLEU.brevity_penalty(candidate, references)
        return bp * math.exp(s)

    def modified_precision(candidate, references, n):

        counts = Counter(ngrams(candidate, n))

        if not counts:
            return 0

        max_counts = {}
        for reference in references:
            reference_counts = Counter(ngrams(reference, n))
            for ngram in counts:
                max_counts[ngram] = max(max_counts.get(ngram, 0), reference_counts[ngram])

        clipped_counts = dict((ngram, min(count, max_counts[ngram])) for ngram, count in counts.items())

        return sum(clipped_counts.values()) / sum(counts.values())

    def brevity_penalty(candidate, references):
        c = len(candidate)
        r = min(abs(len(r) - c) for r in references)

        if c > r:
            return 1
        else:
            return math.exp(1 - r / c)

我想解压缩nltk.bleu_score库，以便能够轻松地将它导入到Android_app开发中。

如果我跑到下面，

bleu = BLEU()    
bleu.compute(candidate, reference_wik)

它返回的错误如下：

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-15-243daadf668f> in <module>()
----> 1 bleu.compute(candidate, reference_wik)

<ipython-input-11-1d4045ea108d> in compute(candidate, references, weights)
      1 class BLEU(object):
      2     def compute(candidate, references, weights):
----> 3         candidate = [c.lower() for c in candidate]
      4         references = [[r.lower() for r in reference] for reference in references]
      5 

TypeError: 'BLEU' object is not iterable

哪里

candidate = ['consists', 'of', 'to', 'make', 'a', 'whole']

reference_wik = [['To', 'be', 'made', 'up', 'of;', 'to', 'consist', 'of', '(especially', 'a', 'comprehensive', 'list', 'of', 'parts).', '[from', 'earlier', '15thc]'], ['To', 'contain', 'or', 'embrace.', '[from', 'earlier', '15thc]'], ['proscribed,', 'usually', 'in', 'the', 'passive)', 'To', 'compose,', 'to', 'constitute.', 'See', 'usage', 'note', 'below'], ['law)', 'To', 'include,', 'contain,', 'or', 'be', 'made', 'up', 'of,', 'defining', 'the', 'minimum', 'elements,', 'whether', 'essential', 'or', 'inessential,', 'to', 'define', 'an', 'invention.', '("Open-ended",', "doesn't", 'limit', 'to', 'the', 'items', 'listed;', 'cf.', 'compose,', 'which', 'is', '"closed"', 'and', 'limits', 'to', 'the', 'items', 'listed)']]

我无法跟踪是什么导致了给定error_message的问题。有调试的提示吗？

Answer 1

任何类函数中的第一个参数是object实例变量，假定它被命名为self，请将其想象为Java中的this。来自官方文档

通常，方法的第一个参数称为self。这不过是一种约定:名称self对Python来说绝对没有什么特殊意义。

从这个开始，回答

方法的第一个参数是调用该方法的实例。这使得方法与函数完全相同，将实际名称留给您使用(尽管self是惯例，当您使用其他东西时，人们通常会对您皱眉)。

因此，将代码更改为：

class BLEU(object):
    # add self in the functions
    def compute(self, candidate, references, weights):
        ...
        # don't do this
        # bp = BLEU.brevity_penalty(candidate, references)
        # instead call the method from like this:
        bp = self.brevity_penalty(candidate, references)
        ...

    def modified_precision(self, candidate, references, n):
        ...

    def brevity_penalty(self, candidate, references):
        ...