[游戏开发]show attend and tell 计算bleu分数（1到4）

Calculate BLEU scores

参考：How to do calculate all bleu scores during evaluvaion #37
主要是参数的改变，默认计算的是BLEU4的分数，从源码中也可以看出来

# Calculate BLEU-4 scores
bleu4 = corpus_bleu(references, hypotheses)
# weights = (1.0 / 1.0,)
bleu1 = corpus_bleu(references, hypotheses, (1.0 / 1.0,))
bleu2 = corpus_bleu(references, hypotheses, (1.0/2.0, 1.0/2.0,))
bleu3 = corpus_bleu(references, hypotheses, (1.0/3.0, 1.0/3.0, 1.0/3.0,))
bleu1_4 = 'bleu1:' + str(bleu1) + '\n' + 'bleu2:' + str(bleu2) + '\n' + 'bleu3:' + str(bleu3) + '\n' + 'bleu4:' + str(bleu4)
print(bleu1_4)

改变bleu4中weights=(0.25, 0.25, 0.25, 0.25)的值即可以得到相应的值

corpus_bleu源码（有兴趣可以全看，没有的话只看参数和注释即可）：

def corpus_bleu(
    list_of_references,
    hypotheses,
    weights=(0.25, 0.25, 0.25, 0.25),
    smoothing_function=None,
    auto_reweigh=False,
):
    """
    Calculate a single corpus-level BLEU score (aka. system-level BLEU) for all
    the hypotheses and their respective references.

    Instead of averaging the sentence level BLEU scores (i.e. macro-average
    precision), the original BLEU metric (Papineni et al. 2002) accounts for
    the micro-average precision (i.e. summing the numerators and denominators
    for each hypothesis-reference(s) pairs before the division).

    >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',
    ...         'ensures', 'that', 'the', 'military', 'always',
    ...         'obeys', 'the', 'commands', 'of', 'the', 'party']
    >>> ref1a = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',
    ...          'ensures', 'that', 'the', 'military', 'will', 'forever',
    ...          'heed', 'Party', 'commands']
    >>> ref1b = ['It', 'is', 'the', 'guiding', 'principle', 'which',
    ...          'guarantees', 'the', 'military', 'forces', 'always',
    ...          'being', 'under', 'the', 'command', 'of', 'the', 'Party']
    >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',
    ...          'army', 'always', 'to', 'heed', 'the', 'directions',
    ...          'of', 'the', 'party']

    >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',
    ...         'interested', 'in', 'world', 'history']
    >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',
    ...          'because', 'he', 'read', 'the', 'book']

    >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]
    >>> hypotheses = [hyp1, hyp2]
    >>> corpus_bleu(list_of_references, hypotheses) # doctest: +ELLIPSIS
    0.5920...

    The example below show that corpus_bleu() is different from averaging
    sentence_bleu() for hypotheses

    >>> score1 = sentence_bleu([ref1a, ref1b, ref1c], hyp1)
    >>> score2 = sentence_bleu([ref2a], hyp2)
    >>> (score1 + score2) / 2 # doctest: +ELLIPSIS
    0.6223...

    :param list_of_references: a corpus of lists of reference sentences, w.r.t. hypotheses
    :type list_of_references: list(list(list(str)))
    :param hypotheses: a list of hypothesis sentences
    :type hypotheses: list(list(str))
    :param weights: weights for unigrams, bigrams, trigrams and so on
    :type weights: list(float)
    :param smoothing_function:
    :type smoothing_function: SmoothingFunction
    :param auto_reweigh: Option to re-normalize the weights uniformly.
    :type auto_reweigh: bool
    :return: The corpus-level BLEU score.
    :rtype: float
    """
    # Before proceeding to compute BLEU, perform sanity checks.

    p_numerators = Counter()  # Key = ngram order, and value = no. of ngram matches.
    p_denominators = Counter()  # Key = ngram order, and value = no. of ngram in ref.
    hyp_lengths, ref_lengths = 0, 0

    assert len(list_of_references) == len(hypotheses), (
        "The number of hypotheses and their reference(s) should be the " "same "
    )

    # Iterate through each hypothesis and their corresponding references.
    for references, hypothesis in zip(list_of_references, hypotheses):
        # For each order of ngram, calculate the numerator and
        # denominator for the corpus-level modified precision.
        for i, _ in enumerate(weights, start=1):
            p_i = modified_precision(references, hypothesis, i)
            p_numerators[i] += p_i.numerator
            p_denominators[i] += p_i.denominator

        # Calculate the hypothesis length and the closest reference length.
        # Adds them to the corpus-level hypothesis and reference counts.
        hyp_len = len(hypothesis)
        hyp_lengths += hyp_len
        ref_lengths += closest_ref_length(references, hyp_len)

    # Calculate corpus-level brevity penalty.
    bp = brevity_penalty(ref_lengths, hyp_lengths)

    # Uniformly re-weighting based on maximum hypothesis lengths if largest
    # order of n-grams < 4 and weights is set at default.
    if auto_reweigh:
        if hyp_lengths < 4 and weights == (0.25, 0.25, 0.25, 0.25):
            weights = (1 / hyp_lengths,) * hyp_lengths

    # Collects the various precision values for the different ngram orders.
    p_n = [
        Fraction(p_numerators[i], p_denominators[i], _normalize=False)
        for i, _ in enumerate(weights, start=1)
    ]

    # Returns 0 if there's no matching n-grams
    # We only need to check for p_numerators[1] == 0, since if there's
    # no unigrams, there won't be any higher order ngrams.
    if p_numerators[1] == 0:
        return 0

    # If there's no smoothing, set use method0 from SmoothinFunction class.
    if not smoothing_function:
        smoothing_function = SmoothingFunction().method0
    # Smoothen the modified precision.
    # Note: smoothing_function() may convert values into floats;
    #       it tries to retain the Fraction object as much as the
    #       smoothing method allows.
    p_n = smoothing_function(
        p_n, references=references, hypothesis=hypothesis, hyp_len=hyp_lengths
    )
    s = (w_i * math.log(p_i) for w_i, p_i in zip(weights, p_n))
    s = bp * math.exp(math.fsum(s))
    return s