031a862a68853167c1be5aa971563f6069fd6d4d,nltk/translate/gleu_score.py,,sentence_gleu,#Any#Any#Any#Any#,16

Before Change

    // For each order of ngram, calculate the no. of ngram matches and
    // keep track of no. of ngram in references.
    ref_ngrams = Counter(everygrams(reference, min_len, max_len))
    hyp_ngrams = Counter(everygrams(hypothesis, min_len, max_len))
    overlap_ngrams = ref_ngrams & hyp_ngrams
    tp = sum(overlap_ngrams.values()) // True positives.
    tpfp = sum(hyp_ngrams.values()) // True positives + False positives.
    tpfn = sum(ref_ngrams.values()) // True positives + False negatives.

    // While defined as the minimum of precision and recall, we can
    // reduce the number of division operations by one by instead finding
    // the maximum of the denominators for the precision and recall
    // formulae, since the numerators are the same:
    //     precision = tp / tpfp
    //     recall = tp / tpfn
    //     min(precision, recall) == tp / max(tpfp, tpfn)

    return tp / max(tpfp, tpfn)


def corpus_gleu(references, hypotheses, min_len=1, max_len=4):
    

After Change

    if not references or isinstance(references[0], string_types):
        references = [references]

    return corpus_gleu(
        [references],
        [hypothesis],
        min_len=min_len,
        max_len=max_len
    )

def corpus_gleu(list_of_references, hypotheses, min_len=1, max_len=4):
    
    Calculate a single corpus-level GLEU score (aka. system-level GLEU) for all

Instances