Add bleu metric (#1834)

* add bleu metric - refactor rouge and add nlp module

* Remove blank line

* autopep8 fix

* Remove metrics

* Add nlp metrics in nlp.__init__

* autopep8 fix

* update

* expose nlp - replace assert by exception

* fix F401

* fix doc

* autopep8 fix

* add test

* Resolve conflict

Co-authored-by: Desroziers <sylvain.desroziers@ifpen.fr>
Co-authored-by: sdesrozis <sdesrozis@users.noreply.github.com>
Co-authored-by: vfdev <vfdev.5@gmail.com>

docs/source/metrics.rst

@@ -325,13 +325,14 @@ Complete list of metrics
     precision.Precision
     PSNR
     recall.Recall
-    Rouge
-    rouge.RougeL
-    rouge.RougeN
     RootMeanSquaredError
     RunningAverage
     SSIM
     TopKCategoricalAccuracy
+    Bleu
+    Rouge
+    RougeL
+    RougeN
 
 Helpers for customizing metrics
 -------------------------------

ignite/metrics/__init__.py

@@ -11,11 +11,12 @@
 from ignite.metrics.metric import BatchFiltered, BatchWise, EpochWise, Metric, MetricUsage
 from ignite.metrics.metrics_lambda import MetricsLambda
 from ignite.metrics.multilabel_confusion_matrix import MultiLabelConfusionMatrix
+from ignite.metrics.nlp.bleu import Bleu
+from ignite.metrics.nlp.rouge import Rouge, RougeL, RougeN
 from ignite.metrics.precision import Precision
 from ignite.metrics.psnr import PSNR
 from ignite.metrics.recall import Recall
 from ignite.metrics.root_mean_squared_error import RootMeanSquaredError
-from ignite.metrics.rouge import Rouge, RougeL, RougeN
 from ignite.metrics.running_average import RunningAverage
 from ignite.metrics.ssim import SSIM
 from ignite.metrics.top_k_categorical_accuracy import TopKCategoricalAccuracy
@@ -43,11 +44,12 @@
     "PSNR",
     "Recall",
     "RootMeanSquaredError",
-    "Rouge",
-    "RougeN",
-    "RougeL",
     "RunningAverage",
     "VariableAccumulation",
     "Frequency",
     "SSIM",
+    "Bleu",
+    "Rouge",
+    "RougeN",
+    "RougeL",
 ]

ignite/metrics/nlp/__init__.py

@@ -0,0 +1,9 @@
+from ignite.metrics.nlp.bleu import Bleu
+from ignite.metrics.nlp.rouge import Rouge, RougeL, RougeN
+
+__all__ = [
+    "Bleu",
+    "Rouge",
+    "RougeN",
+    "RougeL",
+]

ignite/metrics/nlp/bleu.py

@@ -0,0 +1,191 @@
+import math
+from collections import Counter
+from typing import Any, Callable, Sequence, Tuple, Union
+
+import torch
+
+from ignite.exceptions import NotComputableError
+from ignite.metrics.metric import Metric, reinit__is_reduced, sync_all_reduce
+from ignite.metrics.nlp.utils import modified_precision
+
+__all__ = ["Bleu"]
+
+
+def _closest_ref_length(references: Sequence[Sequence[Any]], hyp_len: int) -> int:
+    ref_lens = (len(reference) for reference in references)
+    closest_ref_len = min(ref_lens, key=lambda ref_len: (abs(ref_len - hyp_len), ref_len))
+    return closest_ref_len
+
+
+class _Smoother:
+    """
+    Smoothing helper
+    http://acl2014.org/acl2014/W14-33/pdf/W14-3346.pdf
+    """
+
+    def __init__(self, method: str):
+        valid = ["no_smooth", "smooth1", "nltk_smooth2", "smooth2"]
+        if method not in valid:
+            raise ValueError(f"Smooth is not valid (expected: {valid}, got: {method})")
+        self.smooth = method
+
+    def __call__(self, numerators: Counter, denominators: Counter) -> Sequence[float]:
+        method = getattr(self, self.smooth)
+        return method(numerators, denominators)
+
+    @staticmethod
+    def smooth1(numerators: Counter, denominators: Counter) -> Sequence[float]:
+        epsilon = 0.1
+        denominators_ = [max(1, d) for d in denominators.values()]
+        return [n / d if n != 0 else epsilon / d for n, d in zip(numerators.values(), denominators_)]
+
+    @staticmethod
+    def nltk_smooth2(numerators: Counter, denominators: Counter) -> Sequence[float]:
+        denominators_ = [max(1, d) for d in denominators.values()]
+        return [(n + 1) / (d + 1) for n, d in zip(numerators.values(), denominators_)]
+
+    @staticmethod
+    def smooth2(numerators: Counter, denominators: Counter) -> Sequence[float]:
+        return [(n + 1) / (d + 1) for n, d in zip(numerators.values(), denominators.values())]
+
+    @staticmethod
+    def no_smooth(numerators: Counter, denominators: Counter) -> Sequence[float]:
+        denominators_ = [max(1, d) for d in denominators.values()]
+        return [n / d for n, d in zip(numerators.values(), denominators_)]
+
+
+class Bleu(Metric):
+    r"""Calculates the `BLEU score <https://en.wikipedia.org/wiki/BLEU>`_.
+
+    .. math::
+       \text{BLEU} = b_{p} \cdot \exp \left( \sum_{n=1}^{N} w_{n} \: \log p_{n} \right)
+
+    where :math:`N` is the order of n-grams, :math:`b_{p}` is a sentence brevety penalty, :math:`w_{n}` are
+    positive weights summing to one and :math:`p_{n}` are modified n-gram precisions.
+
+    More details can be found in `Papineni et al. 2002`__.
+
+    __ https://www.aclweb.org/anthology/P02-1040.pdf
+
+    In addition, a review of smoothing techniques can be found in `Chen et al. 2014`__
+
+    __ http://acl2014.org/acl2014/W14-33/pdf/W14-3346.pdf
+
+    Remark :
+
+        This implementation is inspired by nltk
+
+    Args:
+        ngram: order of n-grams.
+        smooth: enable smoothing. Valid are ``no_smooth``, ``smooth1``, ``nltk_smooth2`` or ``smooth2``.
+            Default: ``no_smooth``.
+        output_transform: a callable that is used to transform the
+            :class:`~ignite.engine.engine.Engine`'s ``process_function``'s output into the
+            form expected by the metric. This can be useful if, for example, you have a multi-output model and
+            you want to compute the metric with respect to one of the outputs.
+            By default, metrics require the output as ``(y_pred, y)`` or ``{'y_pred': y_pred, 'y': y}``.
+        device: specifies which device updates are accumulated on. Setting the
+            metric's device to be the same as your ``update`` arguments ensures the ``update`` method is
+            non-blocking. By default, CPU.
+
+    Example:
+
+    .. code-block:: python
+
+        from ignite.metrics.nlp import Bleu
+
+        m = Bleu(ngram=4, smooth="smooth1")
+
+        y_pred = "the the the the the the the"
+        y = ["the cat is on the mat", "there is a cat on the mat"]
+
+        m.update((y_pred.split(), [y.split()]))
+
+        print(m.compute())
+
+    .. versionadded:: 0.5.0
+    """
+
+    def __init__(
+        self,
+        ngram: int = 4,
+        smooth: str = "no_smooth",
+        output_transform: Callable = lambda x: x,
+        device: Union[str, torch.device] = torch.device("cpu"),
+    ):
+        if ngram <= 0:
+            raise ValueError(f"ngram order must be greater than zero (got: {ngram})")
+        self.ngrams_order = ngram
+        self.weights = [1 / self.ngrams_order] * self.ngrams_order
+        self.smoother = _Smoother(method=smooth)
+        super(Bleu, self).__init__(output_transform=output_transform, device=device)
+
+    def _corpus_bleu(self, references: Sequence[Sequence[Any]], candidates: Sequence[Sequence[Any]],) -> float:
+        p_numerators: Counter = Counter()
+        p_denominators: Counter = Counter()
+
+        if len(references) != len(candidates):
+            raise ValueError(
+                f"nb of candidates should be equal to nb of reference lists ({len(candidates)} != "
+                f"{len(references)})"
+            )
+
+        # Iterate through each hypothesis and their corresponding references.
+        for refs, hyp in zip(references, candidates):
+            # For each order of ngram, calculate the numerator and
+            # denominator for the corpus-level modified precision.
+            for i in range(1, self.ngrams_order + 1):
+                numerator, denominator = modified_precision(refs, hyp, i)
+                p_numerators[i] += numerator
+                p_denominators[i] += denominator
+
+        # 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 no smoother, returns 0 if there's at least one a not matching n-grams
+        if self.smoother.smooth == "no_smooth" and min(p_numerators.values()) == 0:
+            return 0
+
+        # Calculate the hypothesis lengths
+        hyp_lengths = [len(hyp) for hyp in candidates]
+
+        # Calculate the closest reference lengths.
+        ref_lengths = [_closest_ref_length(refs, hyp_len) for refs, hyp_len in zip(references, hyp_lengths)]
+
+        # Sum of hypothesis and references lengths
+        hyp_len = sum(hyp_lengths)
+        ref_len = sum(ref_lengths)
+
+        # Calculate corpus-level brevity penalty.
+        if hyp_len < ref_len:
+            bp = math.exp(1 - ref_len / hyp_len) if hyp_len > 0 else 0.0
+        else:
+            bp = 1.0
+
+        # Smoothing
+        p_n = self.smoother(p_numerators, p_denominators)
+
+        # Compute the geometric mean
+        s = [w_i * math.log(p_i) for w_i, p_i in zip(self.weights, p_n)]
+        gm = bp * math.exp(math.fsum(s))
+        return gm
+
+    @reinit__is_reduced
+    def reset(self) -> None:
+        self._sum_of_bleu = torch.tensor(0.0, dtype=torch.double, device=self._device)
+        self._num_sentences = 0
+
+    @reinit__is_reduced
+    def update(self, output: Tuple[Sequence[Any], Sequence[Sequence[Any]]]) -> None:
+        y_pred, y = output
+        self._sum_of_bleu += self._corpus_bleu(references=[y], candidates=[y_pred])
+        self._num_sentences += 1
+
+    @sync_all_reduce("_sum_of_bleu", "_num_sentences")
+    def compute(self) -> torch.Tensor:
+        if self._num_sentences == 0:
+            raise NotComputableError("Bleu must have at least one example before it can be computed.")
+        return self._sum_of_bleu / self._num_sentences

ignite/metrics/rouge.py → ignite/metrics/nlp/rouge.py

@@ -1,5 +1,5 @@
 from abc import ABCMeta, abstractmethod
-from collections import Counter, namedtuple
+from collections import namedtuple
 from typing import Any, Callable, List, Mapping, Optional, Sequence, Tuple, Union
 
 import torch
@@ -9,53 +9,9 @@
 
 # These decorators helps with distributed settings
 from ignite.metrics.metric import reinit__is_reduced, sync_all_reduce
+from ignite.metrics.nlp.utils import lcs, ngrams
 
-
-def ngrams(sequence: Sequence[Any], n: int) -> Counter:
-    """
-    Generate the ngrams from a sequence of items
-
-    Args:
-        sequence: sequence of items
-        n: ngram order
-
-    Returns:
-        A counter of ngram objects
-
-    .. versionadded:: 0.5.0
-    """
-    return Counter([tuple(sequence[i : i + n]) for i in range(len(sequence) - n + 1)])
-
-
-def lcs(seq_a: Sequence[Any], seq_b: Sequence[Any]) -> int:
-    """
-    Compute the length of the longest common subsequence in two sequence of items
-    https://en.wikipedia.org/wiki/Longest_common_subsequence_problem
-
-    Args:
-        seq_a: first sequence of items
-        seq_b: second sequence of items
-
-    Returns:
-        The length of the longest common subsequence
-
-    .. versionadded:: 0.5.0
-    """
-    m = len(seq_a)
-    n = len(seq_b)
-
-    dp = [[0] * (n + 1) for _ in range(m + 1)]
-
-    for i in range(m + 1):
-        for j in range(n + 1):
-            if i == 0 or j == 0:
-                dp[i][j] = 0
-            elif seq_a[i - 1] == seq_b[j - 1]:
-                dp[i][j] = dp[i - 1][j - 1] + 1
-            else:
-                dp[i][j] = max(dp[i - 1][j], dp[i][j - 1])
-
-    return dp[m][n]
+__all__ = ["Rouge", "RougeN", "RougeL"]
 
 
 class Score(namedtuple("Score", ["match", "candidate", "reference"])):
@@ -286,7 +242,7 @@ def __init__(
         super(RougeN, self).__init__(multiref=multiref, alpha=alpha, output_transform=output_transform, device=device)
         self._ngram = ngram
         if self._ngram < 1:
-            raise ValueError(f"ngram order must be greater than one (got : {self._ngram})")
+            raise ValueError(f"ngram order must be greater than zero (got : {self._ngram})")
 
     def _compute_score(self, candidate: Sequence[Any], reference: Sequence[Any]) -> Score:
         return compute_ngram_scores(candidate=candidate, reference=reference, n=self._ngram)