Implement WIL metric

Summary: Implemented functional and class based Word Information Loss metric and added tests.

Reviewed By: ninginthecloud

Differential Revision: D41942828

fbshipit-source-id: 9b1fefce0693344450af80d139ceb3fe5c56279a

tests/metrics/functional/text/test_word_information_lost.py

@@ -0,0 +1,42 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import unittest
+
+import torch
+from torcheval.metrics.functional import word_information_lost
+
+
+class TestWordInformationLost(unittest.TestCase):
+    def test_word_information_lost(self) -> None:
+
+        input = ["hello world", "welcome to the facebook"]
+        target = ["hello metaverse", "welcome to meta"]
+        torch.testing.assert_close(
+            word_information_lost(input, target),
+            torch.tensor(0.7, dtype=torch.float64),
+        )
+
+        input = ["this is the prediction", "there is an other sample"]
+        target = ["this is the reference", "there is another one"]
+        torch.testing.assert_close(
+            word_information_lost(input, target),
+            torch.tensor(0.6527777, dtype=torch.float64),
+        )
+
+    def test_word_information_lost_with_invalid_input(self) -> None:
+        with self.assertRaisesRegex(
+            AssertionError,
+            "Arguments must contain the same number of strings.",
+        ):
+            word_information_lost(
+                ["hello metaverse", "welcome to meta"],
+                [
+                    "welcome to meta",
+                    "this is the prediction",
+                    "there is an other sample",
+                ],
+            )

tests/metrics/text/test_word_information_lost.py

@@ -0,0 +1,49 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from torcheval.metrics.text import WordInformationLost
+from torcheval.utils.test_utils.metric_class_tester import MetricClassTester
+
+
+class TestWordInformationLost(MetricClassTester):
+    def test_word_information_lost(self) -> None:
+        self.run_class_implementation_tests(
+            metric=WordInformationLost(),
+            state_names={"correct_total", "target_total", "preds_total"},
+            update_kwargs={
+                "input": [
+                    ["hello world", "welcome to the facebook"],
+                    ["hello world", "welcome to the facebook"],
+                    ["hello world", "welcome to the facebook"],
+                    ["hello world", "welcome to the facebook"],
+                ],
+                "target": [
+                    ["hello metaverse", "welcome to meta"],
+                    ["hello metaverse", "welcome to meta"],
+                    ["hello metaverse", "welcome to meta"],
+                    ["hello metaverse", "welcome to meta"],
+                ],
+            },
+            compute_result=torch.tensor(0.7, dtype=torch.float64),
+            num_total_updates=4,
+        )
+
+    def test_word_information_lost_with_invalid_input(self) -> None:
+        metric = WordInformationLost()
+
+        with self.assertRaisesRegex(
+            AssertionError,
+            "Arguments must contain the same number of strings.",
+        ):
+            metric.update(
+                ["hello metaverse", "welcome to meta"],
+                [
+                    "welcome to meta",
+                    "this is the prediction",
+                    "there is an other sample",
+                ],
+            )

torcheval/metrics/functional/__init__.py

@@ -44,10 +44,10 @@
 from torcheval.metrics.functional.text import (
     perplexity,
     word_error_rate,
+    word_information_lost,
     word_information_preserved,
 )
 
-
 __all__ = [
     "auc",
     "binary_auprc",
@@ -89,4 +89,5 @@
     "weighted_calibration",
     "word_error_rate",
     "word_information_preserved",
+    "word_information_lost",
 ]

torcheval/metrics/functional/text/__init__.py

@@ -6,8 +6,16 @@
 
 from torcheval.metrics.functional.text.perplexity import perplexity
 from torcheval.metrics.functional.text.word_error_rate import word_error_rate
+from torcheval.metrics.functional.text.word_information_lost import (
+    word_information_lost,
+)
 from torcheval.metrics.functional.text.word_information_preserved import (
     word_information_preserved,
 )
 
-__all__ = ["perplexity", "word_error_rate", "word_information_preserved"]
+__all__ = [
+    "perplexity",
+    "word_error_rate",
+    "word_information_preserved",
+    "word_information_lost",
+]

torcheval/metrics/functional/text/word_information_lost.py

@@ -0,0 +1,76 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from typing import List, Tuple, Union
+
+import torch
+
+from torcheval.metrics.functional.text.helper import _get_errors_and_totals
+
+
+def _wil_update(
+    input: Union[str, List[str]],
+    target: Union[str, List[str]],
+) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+    """Update the wil score with the current set of references and predictions.
+    Args:
+        input: Transcription(s) to score as a string or list of strings
+        target: Reference(s) for each speech input as a string or list of strings
+    Returns:
+        Number of correct words
+        Number of words overall references
+        Number of words overall predictions
+    """
+    if isinstance(input, str):
+        input = [input]
+    if isinstance(target, str):
+        target = [target]
+    assert len(input) == len(
+        target
+    ), f"Arguments must contain the same number of strings, but got len(input)={len(input)} and len(target)={len(target)}"
+    errors, max_total, target_total, input_total = _get_errors_and_totals(input, target)
+    return errors - max_total, target_total, input_total
+
+
+def _wil_compute(
+    correct_total: torch.Tensor, target_total: torch.Tensor, preds_total: torch.Tensor
+) -> torch.Tensor:
+    """Compute the Word Information Lost.
+    Args:
+        correct_total: Number of correct words
+        target_total: Number of words overall references
+        preds_total: Number of words overall prediction
+    Returns:
+        Word Information Lost score
+    """
+    return 1 - ((correct_total / target_total) * (correct_total / preds_total))
+
+
+@torch.inference_mode()
+def word_information_lost(
+    input: Union[str, List[str]],
+    target: Union[str, List[str]],
+) -> torch.Tensor:
+    """Word Information Lost rate is a metric of the performance of an automatic speech recognition system. This
+    value indicates the percentage of characters that were incorrectly predicted. The lower the value, the better
+    the performance of the ASR system with a Word Information Lost rate of 0 being a perfect score.
+
+    Its class version is ``torcheval.metrics.WordInformationLost``.
+
+    Args:
+        input: Transcription(s) to score as a string or list of strings
+        target: Reference(s) for each speech input as a string or list of strings
+    Returns:
+        Word Information Lost rate
+    Examples:
+        >>> from torcheval.metrics.functional import word_information_lost
+        >>> input = ["this is the prediction", "there is an other sample"]
+        >>> target = ["this is the reference", "there is another one"]
+        >>> word_information_lost(input, target)
+        tensor(0.6528)
+    """
+    correct_total, target_total, preds_total = _wil_update(input, target)
+    return _wil_compute(correct_total, target_total, preds_total)

torcheval/metrics/text/__init__.py

@@ -6,6 +6,12 @@
 
 from torcheval.metrics.text.perplexity import Perplexity
 from torcheval.metrics.text.word_error_rate import WordErrorRate
+from torcheval.metrics.text.word_information_lost import WordInformationLost
 from torcheval.metrics.text.word_information_preserved import WordInformationPreserved
 
-__all__ = ["Perplexity", "WordErrorRate", "WordInformationPreserved"]
+__all__ = [
+    "Perplexity",
+    "WordErrorRate",
+    "WordInformationLost",
+    "WordInformationPreserved",
+]

torcheval/metrics/text/word_information_lost.py

@@ -0,0 +1,96 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+# pyre-ignore-all-errors[16]: Undefined attribute of metric states.
+
+from typing import Iterable, List, Optional, TypeVar, Union
+
+import torch
+
+from torcheval.metrics.functional.text.word_information_lost import (
+    _wil_compute,
+    _wil_update,
+)
+
+from torcheval.metrics.metric import Metric
+
+TWordInformationLost = TypeVar("TWordInformationLost")
+
+
+class WordInformationLost(Metric[torch.Tensor]):
+    """Word Information Lost (WIL_) is a metric of the performance of an automatic speech recognition system. This
+    value indicates the percentage of words that were incorrectly predicted between a set of ground-truth sentences
+    and a set of hypothesis sentences. The lower the value, the better the performance of the ASR system with a
+    WordInformationLost of 0 being a perfect score. Word Information Lost rate can then be computed as:
+    .. math::
+        wil = 1 - \frac{C}{N} * \frac{C}{P}
+    where:
+        - :math:`C` is the number of correct words,
+        - :math:`N` is the number of words in the reference
+        - :math:`P` is the number of words in the prediction
+
+    Its functional version is :func:`torcheval.metrics.functional.word_information_lost`.
+
+    Examples:
+        >>> from torcheval.metrics.text import WordInformationLost
+        >>> preds = ["this is the prediction", "there is an other sample"]
+        >>> target = ["this is the reference", "there is another one"]
+        >>> metric = WordInformationLost()
+        >>> metric(preds, target)
+        tensor(0.6528)
+    """
+
+    def __init__(
+        self: TWordInformationLost,
+        device: Optional[torch.device] = None,
+    ) -> None:
+        super().__init__(device=device)
+        self._add_state("correct_total", torch.tensor(0.0, dtype=torch.float64))
+        self._add_state("target_total", torch.tensor(0.0, dtype=torch.float64))
+        self._add_state("preds_total", torch.tensor(0.0, dtype=torch.float64))
+
+    @torch.inference_mode()
+    # pyre-ignore[14]: `update` overrides method defined in `Metric` inconsistently.
+    def update(
+        self: TWordInformationLost,
+        input: Union[str, List[str]],
+        target: Union[str, List[str]],
+    ) -> TWordInformationLost:
+        """Store predictions/references for computing Word Information Lost scores.
+        Args:
+            input: Transcription(s) to score as a string or list of strings
+            target: Reference(s) for each speech input as a string or list of strings
+        """
+        correct_total, target_total, preds_total = _wil_update(input, target)
+        self.correct_total += correct_total
+        self.target_total += target_total
+        self.preds_total += preds_total
+        return self
+
+    @torch.inference_mode()
+    def compute(self: TWordInformationLost) -> torch.Tensor:
+        """Calculate the Word Information Lost.
+        Returns:
+            Word Information Lost score
+        """
+        return _wil_compute(self.correct_total, self.target_total, self.preds_total)
+
+    @torch.inference_mode()
+    def merge_state(
+        self: TWordInformationLost,
+        metrics: Iterable[TWordInformationLost],
+    ) -> TWordInformationLost:
+        """
+        Merge the metric state with its counterparts from other metric instances.
+
+        Args:
+            metrics (Iterable[Metric]): metric instances whose states are to be merged.
+        """
+        for metric in metrics:
+            self.correct_total += metric.correct_total.to(self.device)
+            self.target_total += metric.target_total.to(self.device)
+            self.preds_total += metric.preds_total.to(self.device)
+        return self