Add files

.gitattributes

@@ -53,3 +53,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.jpg filter=lfs diff=lfs merge=lfs -text
 *.jpeg filter=lfs diff=lfs merge=lfs -text
 *.webp filter=lfs diff=lfs merge=lfs -text
+txt filter=lfs diff=lfs merge=lfs -text
+*.txt filter=lfs diff=lfs merge=lfs -text
+*.csv filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,3 +1,32 @@
 ---
-license: mit
+language: fr
+tags:
+- grammar
+- spelling correction
+license: MIT
+datasets:
+- synthetic
 ---
+
+# Spelling correction dataset (French)  
+
+This dataset is generated by transforming/corrupting sentences of a French news corpus
+provided by the [University of Leipzig](https://wortschatz.uni-leipzig.de/en/download/French).
+
+The following transformations are applied to words in the sentences:
+- concatenation of pairs of words
+- swapping of neighboring letters in words
+- insertion
+- deletion
+- replacement (by neighboring characters in AZERTY keyboard)
+
+
+## Generation 
+
+`pip install happytransformer `
+
+```bash
+./scripts/get_data.py -t news -y 2023 -s 10K
+./scripts/generate_dataset.py -i data/fra_news_2023_10k/fra_news_2023_10k-sentences.txt
+```
+

data/dataset/dataset.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:568adf975baad0f56b2a7e366ddfbb66e012cb2fab59de67d326019e2eae2091
+size 13231623

scripts/generate_dataset.py

@@ -0,0 +1,190 @@
+#!/usr/bin/env python3
+
+import sys
+import os
+import argparse
+import random
+import csv
+import re
+import logging
+from tqdm import tqdm
+
+logger = logging.getLogger()
+logger.setLevel(logging.INFO)
+
+
+REPLACEMENT_MAP = {
+    'a': ['&', 'é', 'ã', '1', '2', 'z', 'q'],
+    'ã': ['a', 'é', '&', '"', 'a', 'z', '1', '2', '3'],
+    'à': ['a', 'è', '-', '_', '\\', 'y', 'u', '6', '7', '8'],
+    'ä': ['a', 'â', 'p', 'm', '$', '£', ')', ']', '=', '}', 'º', '+', '%', 'ù'],
+    'â': ['a', 'ä', 'p', 'm', '$', '£', ')', ']', '=', '}', 'º', '+', '%', 'ù'],
+    'b': ['v', 'n', 'g', 'h'],
+    'c': ['x', 'v', 'f', 'g'],
+    'ç': ['c', '_', '\\', 'à', '@', 'i', 'o', '8', '9', '0'],
+    'd': ['s', 'f', 'e', '€', 'r', 'x', 'c'],
+    'e': ['€', 'z', 'r', 's', 'd', '"', '#', '\'', '{', '3', '4'],
+    'é': ['e', '€', 'ã', '&', '"', 'a', 'z', '1', '2', '3'],
+    'ê': ['e', '€', 'z', 'r', 's', 'd', '"', '#', '\'', '{'],
+    'è': ['e', '€', 'à', '-', '_', '\\', 'y', 'u', '6', '7', '8'],
+    'f': ['d', 'g', 'r', 't', 'c', 'v'],
+    'g': ['f', 'h', 't', 'y', 'v', 'b'],
+    'h': ['g', 'j', 'y', 'u', 'b', 'n'],
+    'i': ['u', 'o', '_', '\\', 'ç', '^', 'j', 'k', '8', '9'],
+    'j': ['h', 'k', 'u', 'i', 'n', '?', ','],
+    'k': ['j', 'l', 'i', 'o', '?', ',', '.', ';'],
+    'l': ['k', 'm', 'o', 'p', '.', ';', '/', ':'],
+    'm': ['l', '%', 'ù', 'p', 'ä', 'â', '/', ':', '§', '!'],
+    'n': ['b', '?', ',', 'h', 'j'],
+    'o': ['i', 'p', 'ç', '^', 'à', '@', 'k', 'l', '9', '0'],
+    'p': ['o', 'ä', 'â', 'à', '@', ')', ']', 'l', 'm', '0', 'º'],
+    'q': ['a', 'z', 's', 'w', '>', '<'],
+    'r': ['e', 't', 'd', 'f', '\'', '{', '(', '[', '4', '5'],
+    's': ['q', 'd', 'z', 'w', 'x'],
+    't': ['r', 'y', 'f', 'g', '(', '[', '-', '5', '6'],
+    'u': ['y', 'i', 'h', 'j', 'è', 'à', '_', '\\', '7', '8'],
+    'ù': ['u', '%', 'm', 'ä', 'â', '§', '!', '$', '£'],
+    'v': ['c', 'b', 'f', 'g'],
+    'x': ['w', 'c', 's', 'd'],
+    'y': ['t', 'u', 'g', 'h', '_', 'è', 'à', '6', '7'],
+    'w': ['>', '<', 'x', 'q', 's'],
+    'z': ['a', 'e', '€', 'q', 's', 'é', 'ã', '"', '#', '2', '3'], 
+}
+
+
+SENTENCE_ID = re.compile(r'^\d+\t')
+
+
+def rotate(word: str) -> str:
+    if len(word) < 3:
+        return word
+    i = random.randint(1, len(word) - 2)
+    j = i + 1 if random.random() > 0.5 else i - 1
+    letters = list(word)
+    letters[i], letters[j] = letters[j], letters[i]
+    word_ = ''.join(letters)
+    return word_
+
+
+def replace(word: str) -> str:
+    word_ = '%s' % word
+    i = random.randint(0, len(word) - 1)
+    is_lower = word[i].lower() == word[i]
+    if word[i].lower() in REPLACEMENT_MAP:
+        c = random.choice(REPLACEMENT_MAP[word[i].lower()])
+        c = c if is_lower else c.upper()
+        letters = list(word)
+        letters[i] = c
+        word_ = ''.join(letters)
+    return word_
+
+
+def insert(word: str) -> str:
+    i = random.randint(0, len(word) - 1)
+    c = word[i]
+    j = i + 1 if random.random() > 0.5 else i - 1
+    letters = list(word)
+    letters.insert(j, c)
+    word_ = ''.join(letters)
+    return word_
+
+
+def delete(word: str) -> str:
+    i = random.randint(0, len(word) - 1)
+    word_ = word[:i] + word[i + 1:]
+    return word_
+
+
+OPERATIONS = [
+    rotate,
+    insert,
+    delete,
+    replace
+]
+
+
+def transform_sentence(sentence: str, probability: float = 0.25) -> str:
+    words = sentence.split()
+
+    # Join two words with probability 0.5
+    if random.random() < 0.5 and len(words) > 1:
+        i = random.randint(0, len(words) - 1)
+        if random.random() < 0.5 and i < len(words) - 1 or i == 0:
+            j = i + 1
+            words[i] = f"{words[i]}{words[j]}"
+        else:
+            j = i - 1
+            words[i] = f"{words[j]}{words[i]}"
+        del words[j]
+
+    # Transform each word with probability 'probability'
+    for i, word in enumerate(words):
+        if random.random() < probability and not word.isdigit():
+            words[i] = random.choice(OPERATIONS)(word)
+    sentence_ = ' '.join(words)
+    return sentence_
+
+
+def main(args: argparse.Namespace) -> int:
+    if not os.path.exists(args.input_data_path):
+        logger.error("Invalid input data path.")
+        return -1
+    
+    sentences = []
+    with open(args.input_data_path, 'r') as f:
+        for sentence in f:
+            if SENTENCE_ID.search(sentence):
+                sentence = SENTENCE_ID.sub('', sentence)
+            sentence = (
+                sentence
+                .replace('“', '')
+                .replace('”', '')
+                .replace('"', '')
+                .replace('«', '')
+                .replace('»', '')
+            )   
+            sentences.append(' '.join(sentence.split()))
+
+    dirname = os.path.dirname(args.output_data_path)
+    if not os.path.exists(dirname):
+        os.mkdir(dirname)
+
+    logger.info("Transforming sentences to generate cases")
+    with open(args.output_data_path, 'w', newline='') as csv_file:
+        writer = csv.writer(csv_file, delimiter=',')
+        writer.writerow(["input", "target"])
+        for sentence in tqdm(sentences):
+            correction = sentence
+            for case in range(args.number_of_cases):
+                transformed_sentence = transform_sentence(sentence)
+                if transformed_sentence == sentence:
+                    continue
+                input_text = f"grammaire: {transformed_sentence}"
+                writer.writerow([input_text, correction])
+
+    return 0
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        '--input-data-path', '-i',
+        type=str,
+        required=True
+    )
+    parser.add_argument(
+        '--output-data-path', '-o',
+        type=str,
+        default='data/dataset/dataset.csv'
+    )
+    parser.add_argument(
+        '--number-of-cases', '-n',
+        type=int,
+        default=5
+    )
+
+    args, _ = parser.parse_known_args()
+
+    sys.exit(
+        main(args)
+    )

scripts/get_data.py

@@ -0,0 +1,118 @@
+#!/usr/bin/env python3
+
+import os
+import sys
+import argparse
+import requests
+from enum import Enum
+import urllib.parse as urlparse
+import logging
+import tarfile
+
+
+logger = logging.getLogger()
+logger.setLevel(logging.INFO)
+
+
+class EnumString(Enum):
+    def __str__(self: Enum) -> str:
+        return self.value
+
+
+class CorpusType(EnumString):
+    NEWS = "news"
+    WIKI = "wikipedia"
+
+
+class CorpusSize(EnumString):
+    SMALLEST = "10K"
+    SMALL = "30K"
+    MEDIUM = "100K"
+    LARGE = "300K"
+    LARGEST = "1M"
+
+
+BASE_URL = "https://downloads.wortschatz-leipzig.de"
+URL_PREFIX = "corpora"
+FILE_TEMPLATE = "fra_{type}_{year}_{size}.tar.gz"
+
+
+def main(args: argparse.Namespace) -> int:
+    if not os.path.exists(args.dst_dir):
+        logger.info(f"Invalid destination directory: '{args.dst_dir}'.")
+        return -1
+
+    filename = FILE_TEMPLATE.format(
+        type=args.type.value,
+        year=args.year,
+        size=args.size.value,
+    )
+    url = urlparse.urljoin(
+        BASE_URL,
+        f"{URL_PREFIX}/{filename}"    
+    )
+
+    try:
+        file_path = os.path.join(args.dst_dir, filename)
+        logger.info("Downloading %s" % file_path)
+        with open(file_path, 'wb') as f:
+            response = requests.get(url, stream=True)
+            total_length = response.headers.get('content-length')
+
+            if total_length is None:  # no content length header
+                f.write(response.content)
+            else:
+                dl = 0
+                total_length = int(total_length)
+                for data in response.iter_content(chunk_size=4096):
+                    dl += len(data)
+                    f.write(data)
+                    done = int(50 * dl / total_length)
+                    done_bar = '=' * done
+                    remainder = ' ' * (50 - done)
+                    done_pct = f"{100.0 * dl / total_length:6.2f}"
+                    sys.stdout.write("\r[%s%s] [%s]" % (done_bar, remainder, done_pct))    
+                    sys.stdout.flush()
+    except Exception as error:
+        logger.error(error)
+        return -1
+
+    # Uncompress file
+    try:
+        logger.info("Extracting %s" % file_path)
+        with tarfile.open(file_path) as f:
+            f.extractall(os.path.dirname(file_path))
+    except Exception as error:
+        logger.error(error)
+        return -2
+
+    return 0
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--type", '-t',
+        type=CorpusType,
+        choices=list(CorpusType)
+    )
+    parser.add_argument(
+        "--year", '-y',
+        type=str,
+        default='2023'
+    )
+    parser.add_argument(
+        "--size", '-s',
+        type=CorpusSize,
+        choices=list(CorpusSize)
+    )
+    parser.add_argument(
+        "--dst-dir", '-d',
+        type=str,
+        default='data'
+    )
+    args, _ = parser.parse_known_args()
+    
+    sys.exit(
+        main(args)
+    )