Add script and dataset card

README.md

@@ -0,0 +1,223 @@
+---
+annotations_creators:
+- found
+language_creators:
+- found
+languages:
+- af
+- am
+- ar
+- ast
+- az
+- be
+- bg
+- bn
+- br
+- ca
+- ceb
+- cs
+- cy
+- da
+- de
+- el
+- en
+- eo
+- es
+- et
+- eu
+- fa
+- fi
+- fr
+- fy
+- ga
+- gd
+- gl
+- ha
+- he
+- hi
+- hr
+- hu
+- hy
+- id
+- ig
+- ilo
+- is
+- it
+- ja
+- jv
+- ka
+- kk
+- km
+- ko
+- la
+- lb
+- lg
+- lt
+- lv
+- mg
+- mk
+- ml
+- mr
+- ms
+- my
+- ne
+- nl
+- 'no'
+- oc
+- om
+- or
+- pl
+- pt
+- ro
+- ru
+- sd
+- si
+- sk
+- sl
+- so
+- sq
+- sr
+- su
+- sv
+- sw
+- ta
+- tl
+- tr
+- tt
+- uk
+- ur
+- uz
+- vi
+- wo
+- xh
+- yi
+- yo
+- zh
+- zu
+- se
+licenses:
+- unknown
+multilinguality:
+- multilingual
+size_categories:
+  en-nl:
+  - n<110M
+  en-af:
+  - n<9M
+  en-lt:
+  - <24M
+source_datasets:
+- original
+task_categories:
+- conditional-text-generation
+task_ids:
+- machine-translation
+paperswithcode_id: ccmatrix
+pretty_name: CCMatrixV1
+---
+
+# Dataset Card for CCMatrix v1
+
+## Table of Contents
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
+  - [Languages](#languages)
+- [Dataset Structure](#dataset-structure)
+  - [Data Instances](#data-instances)
+  - [Data Fields](#data-fields)
+  - [Data Splits](#data-splits)
+- [Dataset Creation](#dataset-creation)
+  - [Curation Rationale](#curation-rationale)
+  - [Source Data](#source-data)
+  - [Annotations](#annotations)
+  - [Personal and Sensitive Information](#personal-and-sensitive-information)
+- [Considerations for Using the Data](#considerations-for-using-the-data)
+  - [Social Impact of Dataset](#social-impact-of-dataset)
+  - [Discussion of Biases](#discussion-of-biases)
+  - [Other Known Limitations](#other-known-limitations)
+- [Additional Information](#additional-information)
+  - [Dataset Curators](#dataset-curators)
+  - [Licensing Information](#licensing-information)
+  - [Citation Information](#citation-information)
+  - [Contributions](#contributions)
+
+## Dataset Description
+- **Homepage:** https://opus.nlpl.eu/CCMatrix.php
+- **Repository:** None
+- **Paper:** https://arxiv.org/abs/1911.04944
+### Dataset Summary
+
+This corpus has been extracted from web crawls using the margin-based bitext mining techniques described at https://github.com/facebookresearch/LASER/tree/master/tasks/CCMatrix.
+
+* 90 languages, 1,197 bitexts
+* total number of files: 90
+* total number of tokens: 112.14G
+* total number of sentence fragments: 7.37G
+
+### Supported Tasks and Leaderboards
+[More Information Needed]
+### Languages
+
+To load a language pair which isn't part of the config, all you need to do is specify the language code as pairs.
+You can find the valid pairs in Homepage section of Dataset Description: https://opus.nlpl.eu/CCMatrix.php
+E.g.
+
+```
+dataset = load_dataset("yhavinga/ccmatrix", lang1="en", lang2="nl")
+```
+
+## Dataset Structure
+### Data Instances
+For example:
+
+```json
+{
+        "id": 1,
+        "score": 1.2498379,
+        "translation": {
+            "nl": "En we moeten elke waarheid vals noemen die niet minstens door een lach vergezeld ging.”",
+            "en": "And we should call every truth false which was not accompanied by at least one laugh.”"
+        }
+    }
+```
+
+### Data Fields
+Each example contains an integer id starting with 0, a score, and a translation dictionary with the language 1 and
+language 2 texts.
+
+### Data Splits
+Only a `train` split is provided.
+
+## Dataset Creation
+### Curation Rationale
+[More Information Needed]
+### Source Data
+[More Information Needed]
+#### Initial Data Collection and Normalization
+[More Information Needed]
+#### Who are the source language producers?
+[More Information Needed]
+### Annotations
+[More Information Needed]
+#### Annotation process
+[More Information Needed]
+#### Who are the annotators?
+[More Information Needed]
+### Personal and Sensitive Information
+[More Information Needed]
+## Considerations for Using the Data
+### Social Impact of Dataset
+[More Information Needed]
+### Discussion of Biases
+[More Information Needed]
+### Other Known Limitations
+[More Information Needed]
+## Additional Information
+### Dataset Curators
+[More Information Needed]
+### Licensing Information
+[More Information Needed]
+### Citation Information
+[More Information Needed]
+### Contributions

ccmatrix.py

@@ -0,0 +1,146 @@
+# coding=utf-8
+# Copyright 2020 HuggingFace Datasets 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.
+
+# Lint as: python3
+import os
+
+import datasets
+
+_DESCRIPTION = """\
+CCMatrix: Mining Billions of High-Quality Parallel Sentences on the WEB
+
+We show that margin-based bitext mining in LASER's multilingual sentence space can be applied to
+monolingual corpora of billions of sentences to produce high quality aligned translation data.
+We use thirty-two snapshots of a curated common crawl corpus [1] totaling 69 billion unique sentences.
+Using one unified approach for 80 languages, we were able to mine 10.8 billion parallel sentences,
+out of which only 2.9 billion are aligned with English.
+
+IMPORTANT: Please cite reference [2][3] if you use this data.
+
+[1] Guillaume Wenzek, Marie-Anne Lachaux, Alexis Conneau, Vishrav Chaudhary, Francisco Guzmán, Armand Jouli
+    and Edouard Grave, CCNet: Extracting High Quality Monolingual Datasets from Web Crawl Data
+
+[2] Holger Schwenk, Guillaume Wenzek, Sergey Edunov, Edouard Grave and Armand Joulin,
+    CCMatrix: Mining Billions of High-Quality Parallel Sentences on the WEB
+
+[3] Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines,
+    Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky,
+    Sergey Edunov, Edouard Grave, Michael Auli, and Armand Joulin.
+    Beyond English-Centric Multilingual Machine Translation
+    
+90 languages, 1,197 bitexts
+total number of files: 90
+total number of tokens: 112.14G
+total number of sentence fragments: 7.37G
+"""
+_HOMEPAGE_URL = "https://opus.nlpl.eu/CCMatrix.php"
+_CITATION = """\
+ Guillaume Wenzek, Marie-Anne Lachaux, Alexis Conneau, Vishrav Chaudhary, Francisco Guzmán, Armand Jouli and Edouard Grave, CCNet: Extracting High Quality Monolingual Datasets from Web Crawl Data
+"""
+
+_VERSION = "1.0.0"
+_FILE = "CCMatrix.{}.{}"  # E.g. CCMatrix.en-nl.nl
+_DOWNLOAD_URL = "https://opus.nlpl.eu/download.php?f=CCMatrix/v1/moses/{}.txt.zip"
+
+_LANGUAGES = ["nl", "en", "de", "fr", "es", "lt", "it"]
+
+_LANGUAGE_PAIRS = [(l1, l2) for l1 in _LANGUAGES for l2 in _LANGUAGES if l1 != l2]
+_SIZES = ["", "1000_000", "25_000_000"]
+
+_CONFIGS = [(l1, l2, size) for (l1, l2) in _LANGUAGE_PAIRS for size in _SIZES]
+
+
+class CCMatrixConfig(datasets.BuilderConfig):
+    def __init__(self, *args, lang1=None, lang2=None, size=None, **kwargs):
+        super().__init__(
+            *args,
+            name=f"{lang1}-{lang2}{'-' + size if size else ''}",
+            **kwargs,
+        )
+        self.lang1 = lang1
+        self.lang2 = lang2
+        self.size = size
+        x, y = (lang1, lang2) if lang1 < lang2 else (lang2, lang1)
+        self.download_pair = f"{x}-{y}"
+
+
+class CCMatrix(datasets.GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [
+        CCMatrixConfig(
+            lang1=lang1,
+            lang2=lang2,
+            size=size,
+            description=f"Translating {lang1} to {lang2} or vice versa{ ' ' + size + ' rows' if size else ''}",
+            version=datasets.Version(_VERSION),
+        )
+        for lang1, lang2, size in _CONFIGS
+    ]
+    BUILDER_CONFIG_CLASS = CCMatrixConfig
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "id": datasets.Value("int32"),
+                    "score": datasets.Value("float"),
+                    "translation": datasets.Translation(
+                        languages=(self.config.lang1, self.config.lang2)
+                    ),
+                },
+            ),
+            supervised_keys=None,
+            homepage=_HOMEPAGE_URL,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        download_url = _DOWNLOAD_URL.format(self.config.download_pair)
+        path = dl_manager.download_and_extract(download_url)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={"datapath": path},
+            )
+        ]
+
+    def _generate_examples(self, datapath):
+        l1_path = os.path.join(
+            datapath, _FILE.format(self.config.download_pair, self.config.lang1)
+        )
+        l2_path = os.path.join(
+            datapath, _FILE.format(self.config.download_pair, self.config.lang2)
+        )
+        scores_path = os.path.join(
+            datapath, _FILE.format(self.config.download_pair, "scores")
+        )
+        with open(l1_path, encoding="utf-8") as f1, open(
+            l2_path, encoding="utf-8"
+        ) as f2, open(scores_path, encoding="utf-8") as f3:
+            for sentence_counter, (x, y, score) in enumerate(zip(f1, f2, f3)):
+                if self.config.size and sentence_counter == int(self.config.size):
+                    return
+                result = (
+                    sentence_counter,
+                    {
+                        "id": sentence_counter,
+                        "score": score,
+                        "translation": {
+                            self.config.lang1: x.strip(),
+                            self.config.lang2: y.strip(),
+                        },
+                    },
+                )
+                yield result

test_ccmatrix.py

@@ -0,0 +1,108 @@
+from datasets import load_dataset
+
+
+def test_streaming_dataset():
+    datasets = load_dataset("./ccmatrix.py", lang1="nl", lang2="en", streaming=True)
+    assert list(datasets.keys()) == ["train"]
+
+    train_ds = datasets["train"]
+
+    i = iter(train_ds)
+    e = next(i)
+
+    assert e == {
+        "id": 0,
+        "score": 1.2499677,
+        "translation": {
+            "nl": "Zij kwamen uit alle delen van Egypte, evenals zij op de dag van Zijn komst zullen doen.",
+            "en": "They come from all parts of Egypt, just like they will at the day of His coming.",
+        },
+    }
+
+    e = next(i)
+
+    assert list(e.keys()) == ["id", "score", "translation"]
+
+    assert e == {
+        "id": 1,
+        "score": 1.2498379,
+        "translation": {
+            "nl": "En we moeten elke waarheid vals noemen die niet minstens door een lach vergezeld ging.”",
+            "en": 'And we should call every truth false which was not accompanied by at least one laugh."',
+        },
+    }
+
+
+def test_streaming_dataset_2():
+    datasets = load_dataset("./ccmatrix.py", "nl-en", streaming=True)
+    assert list(datasets.keys()) == ["train"]
+
+    train_ds = datasets["train"]
+
+    i = iter(train_ds)
+    e = next(i)
+
+    assert e == {
+        "id": 0,
+        "score": 1.2499677,
+        "translation": {
+            "nl": "Zij kwamen uit alle delen van Egypte, evenals zij op de dag van Zijn komst zullen doen.",
+            "en": "They come from all parts of Egypt, just like they will at the day of His coming.",
+        },
+    }
+
+    e = next(i)
+
+    assert list(e.keys()) == ["id", "score", "translation"]
+
+    assert e == {
+        "id": 1,
+        "score": 1.2498379,
+        "translation": {
+            "nl": "En we moeten elke waarheid vals noemen die niet minstens door een lach vergezeld ging.”",
+            "en": 'And we should call every truth false which was not accompanied by at least one laugh."',
+        },
+    }
+
+
+def test_small_config():
+    datasets = load_dataset("./ccmatrix.py", "nl-en-1000_000")
+    assert list(datasets.keys()) == ["train"]
+
+    train_ds = datasets["train"]
+    assert len(train_ds) == 1000000
+
+    i = iter(train_ds)
+    e = next(i)
+
+    assert e == {
+        "id": 0,
+        "score": 1.2499676942825317,
+        "translation": {
+            "nl": "Zij kwamen uit alle delen van Egypte, evenals zij op de dag van Zijn komst zullen doen.",
+            "en": "They come from all parts of Egypt, just like they will at the day of His coming.",
+        },
+    }
+
+    e = next(i)
+
+    assert list(e.keys()) == ["id", "score", "translation"]
+
+    assert e == {
+        "id": 1,
+        "score": 1.249837875366211,
+        "translation": {
+            "nl": "En we moeten elke waarheid vals noemen die niet minstens door een lach vergezeld ging.”",
+            "en": 'And we should call every truth false which was not accompanied by at least one laugh."',
+        },
+    }
+
+
+def test_medium_config():
+    datasets = load_dataset("./ccmatrix.py", "nl-en-25_000_000", streaming=True)
+    assert list(datasets.keys()) == ["train"]
+
+    train_ds = datasets["train"]
+
+    i = iter(train_ds)
+    e = next(i)