Update Space (evaluate main: 828c6327)

README.md

@@ -1,12 +1,147 @@
 ---
-title: Chrf
-emoji: ⚡
-colorFrom: purple
-colorTo: indigo
+title: chrF
+emoji: 🤗 
+colorFrom: blue
+colorTo: red
 sdk: gradio
 sdk_version: 3.0.2
 app_file: app.py
 pinned: false
+tags:
+- evaluate
+- metric
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces#reference
+# Metric Card for chrF(++)
+
+
+## Metric Description
+ChrF and ChrF++ are two MT evaluation metrics that use the F-score statistic for character n-gram matches. ChrF++ additionally includes word n-grams, which correlate more strongly with direct assessment. We use the implementation that is already present in sacrebleu.
+
+While this metric is included in sacreBLEU, the implementation here is slightly different from sacreBLEU in terms of the required input format. Here, the length of the references and hypotheses lists need to be the same, so you may need to transpose your references compared to sacrebleu's required input format. See https://github.com/huggingface/datasets/issues/3154#issuecomment-950746534
+
+See the [sacreBLEU README.md](https://github.com/mjpost/sacreBLEU#chrf--chrf) for more information.
+
+
+## How to Use
+At minimum, this metric requires a `list` of predictions and a `list` of `list`s of references:
+```python
+>>> prediction = ["The relationship between cats and dogs is not exactly friendly.", "a good bookshop is just a genteel black hole that knows how to read."]
+>>> reference = [["The relationship between dogs and cats is not exactly friendly.", ], ["A good bookshop is just a genteel Black Hole that knows how to read."]]
+>>> chrf = evaluate.load("chrf")
+>>> results = chrf.compute(predictions=prediction, references=reference)
+>>> print(results)
+{'score': 84.64214891738334, 'char_order': 6, 'word_order': 0, 'beta': 2}
+```
+
+### Inputs
+- **`predictions`** (`list` of `str`): The predicted sentences.
+- **`references`** (`list` of `list` of `str`): The references. There should be one reference sub-list for each prediction sentence.
+- **`char_order`** (`int`): Character n-gram order. Defaults to `6`. 
+- **`word_order`** (`int`): Word n-gram order. If equals to 2, the metric is referred to as chrF++. Defaults to `0`.
+- **`beta`** (`int`): Determine the importance of recall w.r.t precision. Defaults to `2`.
+- **`lowercase`** (`bool`): If `True`, enables case-insensitivity. Defaults to `False`.
+- **`whitespace`** (`bool`): If `True`, include whitespaces when extracting character n-grams. Defaults to `False`.
+- **`eps_smoothing`** (`bool`): If `True`, applies epsilon smoothing similar to reference chrF++.py, NLTK, and Moses implementations. If `False`, takes into account effective match order similar to sacreBLEU < 2.0.0. Defaults to `False`.
+
+
+
+### Output Values
+The output is a dictionary containing the following fields:
+- **`'score'`** (`float`): The chrF (chrF++) score.
+- **`'char_order'`** (`int`): The character n-gram order.
+- **`'word_order'`** (`int`): The word n-gram order. If equals to `2`, the metric is referred to as chrF++.
+- **`'beta'`** (`int`): Determine the importance of recall w.r.t precision.
+
+
+The output is formatted as below:
+```python
+{'score': 61.576379378113785, 'char_order': 6, 'word_order': 0, 'beta': 2}
+```
+
+The chrF(++) score can be any value between `0.0` and `100.0`, inclusive.
+
+#### Values from Popular Papers
+
+
+### Examples
+A simple example of calculating chrF:
+```python
+>>> prediction = ["The relationship between cats and dogs is not exactly friendly.", "a good bookshop is just a genteel black hole that knows how to read."]
+>>> reference = [["The relationship between dogs and cats is not exactly friendly.", ], ["A good bookshop is just a genteel Black Hole that knows how to read."]]
+>>> chrf = evaluate.load("chrf")
+>>> results = chrf.compute(predictions=prediction, references=reference)
+>>> print(results)
+{'score': 84.64214891738334, 'char_order': 6, 'word_order': 0, 'beta': 2}
+```
+
+The same example, but with the argument `word_order=2`, to calculate chrF++ instead of chrF:
+```python
+>>> prediction = ["The relationship between cats and dogs is not exactly friendly.", "a good bookshop is just a genteel black hole that knows how to read."]
+>>> reference = [["The relationship between dogs and cats is not exactly friendly.", ], ["A good bookshop is just a genteel Black Hole that knows how to read."]]
+>>> chrf = evaluate.load("chrf")
+>>> results = chrf.compute(predictions=prediction,
+...                         references=reference,
+...                         word_order=2)
+>>> print(results)
+{'score': 82.87263732906315, 'char_order': 6, 'word_order': 2, 'beta': 2}
+```
+
+The same chrF++ example as above, but with `lowercase=True` to normalize all case:
+```python
+>>> prediction = ["The relationship between cats and dogs is not exactly friendly.", "a good bookshop is just a genteel black hole that knows how to read."]
+>>> reference = [["The relationship between dogs and cats is not exactly friendly.", ], ["A good bookshop is just a genteel Black Hole that knows how to read."]]
+>>> chrf = evaluate.load("chrf")
+>>> results = chrf.compute(predictions=prediction,
+...                         references=reference,
+...                         word_order=2,
+...                         lowercase=True)
+>>> print(results)
+{'score': 92.12853119829202, 'char_order': 6, 'word_order': 2, 'beta': 2}
+```
+
+
+## Limitations and Bias
+- According to [Popović 2017](https://www.statmt.org/wmt17/pdf/WMT70.pdf), chrF+ (where `word_order=1`) and chrF++ (where `word_order=2`) produce scores that correlate better with human judgements than chrF (where `word_order=0`) does. 
+
+## Citation
+```bibtex
+@inproceedings{popovic-2015-chrf,
+    title = "chr{F}: character n-gram {F}-score for automatic {MT} evaluation",
+    author = "Popovi{\'c}, Maja",
+    booktitle = "Proceedings of the Tenth Workshop on Statistical Machine Translation",
+    month = sep,
+    year = "2015",
+    address = "Lisbon, Portugal",
+    publisher = "Association for Computational Linguistics",
+    url = "https://aclanthology.org/W15-3049",
+    doi = "10.18653/v1/W15-3049",
+    pages = "392--395",
+}
+@inproceedings{popovic-2017-chrf,
+    title = "chr{F}++: words helping character n-grams",
+    author = "Popovi{\'c}, Maja",
+    booktitle = "Proceedings of the Second Conference on Machine Translation",
+    month = sep,
+    year = "2017",
+    address = "Copenhagen, Denmark",
+    publisher = "Association for Computational Linguistics",
+    url = "https://aclanthology.org/W17-4770",
+    doi = "10.18653/v1/W17-4770",
+    pages = "612--618",
+}
+@inproceedings{post-2018-call,
+    title = "A Call for Clarity in Reporting {BLEU} Scores",
+    author = "Post, Matt",
+    booktitle = "Proceedings of the Third Conference on Machine Translation: Research Papers",
+    month = oct,
+    year = "2018",
+    address = "Belgium, Brussels",
+    publisher = "Association for Computational Linguistics",
+    url = "https://www.aclweb.org/anthology/W18-6319",
+    pages = "186--191",
+}
+```
+
+## Further References
+- See the [sacreBLEU README.md](https://github.com/mjpost/sacreBLEU#chrf--chrf) for more information on this implementation.

app.py

@@ -0,0 +1,6 @@
+import evaluate
+from evaluate.utils import launch_gradio_widget
+
+
+module = evaluate.load("chrf")
+launch_gradio_widget(module)

chrf.py

@@ -0,0 +1,175 @@
+# Copyright 2021 The HuggingFace Evaluate Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" Chrf(++) metric as available in sacrebleu. """
+import datasets
+import sacrebleu as scb
+from packaging import version
+from sacrebleu import CHRF
+
+import evaluate
+
+
+_CITATION = """\
+@inproceedings{popovic-2015-chrf,
+    title = "chr{F}: character n-gram {F}-score for automatic {MT} evaluation",
+    author = "Popovi{\'c}, Maja",
+    booktitle = "Proceedings of the Tenth Workshop on Statistical Machine Translation",
+    month = sep,
+    year = "2015",
+    address = "Lisbon, Portugal",
+    publisher = "Association for Computational Linguistics",
+    url = "https://aclanthology.org/W15-3049",
+    doi = "10.18653/v1/W15-3049",
+    pages = "392--395",
+}
+@inproceedings{popovic-2017-chrf,
+    title = "chr{F}++: words helping character n-grams",
+    author = "Popovi{\'c}, Maja",
+    booktitle = "Proceedings of the Second Conference on Machine Translation",
+    month = sep,
+    year = "2017",
+    address = "Copenhagen, Denmark",
+    publisher = "Association for Computational Linguistics",
+    url = "https://aclanthology.org/W17-4770",
+    doi = "10.18653/v1/W17-4770",
+    pages = "612--618",
+}
+@inproceedings{post-2018-call,
+    title = "A Call for Clarity in Reporting {BLEU} Scores",
+    author = "Post, Matt",
+    booktitle = "Proceedings of the Third Conference on Machine Translation: Research Papers",
+    month = oct,
+    year = "2018",
+    address = "Belgium, Brussels",
+    publisher = "Association for Computational Linguistics",
+    url = "https://www.aclweb.org/anthology/W18-6319",
+    pages = "186--191",
+}
+"""
+
+_DESCRIPTION = """\
+ChrF and ChrF++ are two MT evaluation metrics. They both use the F-score statistic for character n-gram matches,
+and ChrF++ adds word n-grams as well which correlates more strongly with direct assessment. We use the implementation
+that is already present in sacrebleu.
+
+The implementation here is slightly different from sacrebleu in terms of the required input format. The length of
+the references and hypotheses lists need to be the same, so you may need to transpose your references compared to
+sacrebleu's required input format. See https://github.com/huggingface/datasets/issues/3154#issuecomment-950746534
+
+See the README.md file at https://github.com/mjpost/sacreBLEU#chrf--chrf for more information.
+"""
+
+_KWARGS_DESCRIPTION = """
+Produces ChrF(++) scores for hypotheses given reference translations.
+
+Args:
+    predictions (list of str): The predicted sentences.
+    references (list of list of str): The references. There should be one reference sub-list for each prediction sentence.
+    char_order (int): Character n-gram order. Defaults to `6`.
+    word_order (int): Word n-gram order. If equals to `2`, the metric is referred to as chrF++. Defaults to `0`.
+    beta (int): Determine the importance of recall w.r.t precision. Defaults to `2`.
+    lowercase (bool): if `True`, enables case-insensitivity. Defaults to `False`.
+    whitespace (bool): If `True`, include whitespaces when extracting character n-grams.
+    eps_smoothing (bool): If `True`, applies epsilon smoothing similar
+    to reference chrF++.py, NLTK and Moses implementations. If `False`,
+    it takes into account effective match order similar to sacreBLEU < 2.0.0. Defaults to `False`.
+
+Returns:
+    'score' (float): The chrF (chrF++) score,
+    'char_order' (int): The character n-gram order,
+    'word_order' (int): The word n-gram order. If equals to 2, the metric is referred to as chrF++,
+    'beta' (int): Determine the importance of recall w.r.t precision
+
+Examples:
+    Example 1--a simple example of calculating chrF:
+        >>> prediction = ["The relationship between cats and dogs is not exactly friendly.", "a good bookshop is just a genteel black hole that knows how to read."]
+        >>> reference = [["The relationship between dogs and cats is not exactly friendly."], ["A good bookshop is just a genteel Black Hole that knows how to read."]]
+        >>> chrf = evaluate.load("chrf")
+        >>> results = chrf.compute(predictions=prediction, references=reference)
+        >>> print(results)
+        {'score': 84.64214891738334, 'char_order': 6, 'word_order': 0, 'beta': 2}
+
+    Example 2--the same example, but with the argument word_order=2, to calculate chrF++ instead of chrF:
+        >>> prediction = ["The relationship between cats and dogs is not exactly friendly.", "a good bookshop is just a genteel black hole that knows how to read."]
+        >>> reference = [["The relationship between dogs and cats is not exactly friendly."], ["A good bookshop is just a genteel Black Hole that knows how to read."]]
+        >>> chrf = evaluate.load("chrf")
+        >>> results = chrf.compute(predictions=prediction,
+        ...                         references=reference,
+        ...                         word_order=2)
+        >>> print(results)
+        {'score': 82.87263732906315, 'char_order': 6, 'word_order': 2, 'beta': 2}
+
+    Example 3--the same chrF++ example as above, but with `lowercase=True` to normalize all case:
+        >>> prediction = ["The relationship between cats and dogs is not exactly friendly.", "a good bookshop is just a genteel black hole that knows how to read."]
+        >>> reference = [["The relationship between dogs and cats is not exactly friendly."], ["A good bookshop is just a genteel Black Hole that knows how to read."]]
+        >>> chrf = evaluate.load("chrf")
+        >>> results = chrf.compute(predictions=prediction,
+        ...                         references=reference,
+        ...                         word_order=2,
+        ...                         lowercase=True)
+        >>> print(results)
+        {'score': 92.12853119829202, 'char_order': 6, 'word_order': 2, 'beta': 2}
+"""
+
+
+@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
+class ChrF(evaluate.EvaluationModule):
+    def _info(self):
+        if version.parse(scb.__version__) < version.parse("1.4.12"):
+            raise ImportWarning(
+                "To use `sacrebleu`, the module `sacrebleu>=1.4.12` is required, and the current version of `sacrebleu` doesn't match this condition.\n"
+                'You can install it with `pip install "sacrebleu>=1.4.12"`.'
+            )
+        return evaluate.EvaluationModuleInfo(
+            description=_DESCRIPTION,
+            citation=_CITATION,
+            homepage="https://github.com/mjpost/sacreBLEU#chrf--chrf",
+            inputs_description=_KWARGS_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "predictions": datasets.Value("string", id="sequence"),
+                    "references": datasets.Sequence(datasets.Value("string", id="sequence"), id="references"),
+                }
+            ),
+            codebase_urls=["https://github.com/mjpost/sacreBLEU#chrf--chrf"],
+            reference_urls=[
+                "https://github.com/m-popovic/chrF",
+            ],
+        )
+
+    def _compute(
+        self,
+        predictions,
+        references,
+        char_order: int = CHRF.CHAR_ORDER,
+        word_order: int = CHRF.WORD_ORDER,
+        beta: int = CHRF.BETA,
+        lowercase: bool = False,
+        whitespace: bool = False,
+        eps_smoothing: bool = False,
+    ):
+        references_per_prediction = len(references[0])
+        if any(len(refs) != references_per_prediction for refs in references):
+            raise ValueError("Sacrebleu requires the same number of references for each prediction")
+        transformed_references = [[refs[i] for refs in references] for i in range(references_per_prediction)]
+
+        sb_chrf = CHRF(char_order, word_order, beta, lowercase, whitespace, eps_smoothing)
+        output = sb_chrf.corpus_score(predictions, transformed_references)
+
+        return {
+            "score": output.score,
+            "char_order": output.char_order,
+            "word_order": output.word_order,
+            "beta": output.beta,
+        }

requirements.txt

@@ -0,0 +1,4 @@
+# TODO: fix github to release
+git+https://github.com/huggingface/evaluate.git@b6e6ed7f3e6844b297bff1b43a1b4be0709b9671
+datasets~=2.0
+sacrebleu