Refactor with dataclass (#8)
Browse files* use dataclass
* update README
- README.md +20 -0
- livedoor-news-corpus.py +22 -40
README.md
CHANGED
@@ -1,3 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
# Dataset Card for Livedoor News Corpus
|
2 |
|
3 |
[![CI](https://github.com/shunk031/huggingface-datasets_livedoor-news-corpus/actions/workflows/ci.yaml/badge.svg)](https://github.com/shunk031/huggingface-datasets_livedoor-news-corpus/actions/workflows/ci.yaml)
|
|
|
1 |
+
---
|
2 |
+
annotations_creators: []
|
3 |
+
language:
|
4 |
+
- ja
|
5 |
+
language_creators:
|
6 |
+
- found
|
7 |
+
license:
|
8 |
+
- cc-by-nd-4.0
|
9 |
+
multilinguality:
|
10 |
+
- monolingual
|
11 |
+
pretty_name: livedoor-news-corpus
|
12 |
+
size_categories: []
|
13 |
+
source_datasets: []
|
14 |
+
tags: []
|
15 |
+
task_categories:
|
16 |
+
- text-classification
|
17 |
+
task_ids:
|
18 |
+
- multi-class-classification
|
19 |
+
---
|
20 |
+
|
21 |
# Dataset Card for Livedoor News Corpus
|
22 |
|
23 |
[![CI](https://github.com/shunk031/huggingface-datasets_livedoor-news-corpus/actions/workflows/ci.yaml/badge.svg)](https://github.com/shunk031/huggingface-datasets_livedoor-news-corpus/actions/workflows/ci.yaml)
|
livedoor-news-corpus.py
CHANGED
@@ -1,9 +1,11 @@
|
|
1 |
import logging
|
|
|
2 |
import pathlib
|
3 |
-
from typing import Dict, List, Union, Optional
|
4 |
import random
|
|
|
|
|
|
|
5 |
import datasets as ds
|
6 |
-
import math
|
7 |
|
8 |
logger = logging.getLogger(__name__)
|
9 |
|
@@ -25,45 +27,24 @@ _LICENSE = """\
|
|
25 |
_DOWNLOAD_URL = "https://www.rondhuit.com/download/ldcc-20140209.tar.gz"
|
26 |
|
27 |
|
|
|
28 |
class LivedoorNewsCorpusConfig(ds.BuilderConfig):
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
shuffle: bool = False,
|
35 |
-
random_state: int = 0,
|
36 |
-
name: str = "default",
|
37 |
-
version: Optional[Union[ds.utils.Version, str]] = ds.utils.Version("0.0.0"),
|
38 |
-
data_dir: Optional[str] = None,
|
39 |
-
data_files: Optional[ds.data_files.DataFilesDict] = None,
|
40 |
-
description: Optional[str] = None,
|
41 |
-
) -> None:
|
42 |
-
super().__init__(
|
43 |
-
name=name,
|
44 |
-
version=version,
|
45 |
-
data_dir=data_dir,
|
46 |
-
data_files=data_files,
|
47 |
-
description=description,
|
48 |
-
)
|
49 |
-
assert train_ratio + val_ratio + test_ratio == 1.0
|
50 |
-
|
51 |
-
self.train_ratio = train_ratio
|
52 |
-
self.val_ratio = val_ratio
|
53 |
-
self.test_ratio = test_ratio
|
54 |
|
55 |
-
|
56 |
-
self.
|
57 |
|
58 |
|
59 |
class LivedoorNewsCorpusDataset(ds.GeneratorBasedBuilder):
|
60 |
-
VERSION = ds.Version("1.0.0")
|
61 |
-
|
62 |
-
BUILDER_CONFIG_CLASS = LivedoorNewsCorpusConfig # type: ignore
|
63 |
-
|
64 |
BUILDER_CONFIGS = [
|
65 |
LivedoorNewsCorpusConfig(
|
66 |
-
version=VERSION,
|
67 |
description="Livedoor ニュースコーパス",
|
68 |
)
|
69 |
]
|
@@ -105,14 +86,16 @@ class LivedoorNewsCorpusDataset(ds.GeneratorBasedBuilder):
|
|
105 |
article_paths = list(dataset_root_dir.glob("*/**/*.txt"))
|
106 |
article_paths = list(filter(lambda p: p.name != "LICENSE.txt", article_paths))
|
107 |
|
108 |
-
|
109 |
-
|
|
|
|
|
110 |
random.shuffle(article_paths)
|
111 |
|
112 |
num_articles = len(article_paths)
|
113 |
-
num_tng = math.ceil(num_articles *
|
114 |
-
num_val = math.ceil(num_articles *
|
115 |
-
num_tst = math.ceil(num_articles *
|
116 |
|
117 |
tng_articles = article_paths[:num_tng]
|
118 |
val_articles = article_paths[num_tng : num_tng + num_val]
|
@@ -150,7 +133,6 @@ class LivedoorNewsCorpusDataset(ds.GeneratorBasedBuilder):
|
|
150 |
return example_dict
|
151 |
|
152 |
def _generate_examples(self, article_paths: List[pathlib.Path]): # type: ignore[override]
|
153 |
-
|
154 |
for i, article_path in enumerate(article_paths):
|
155 |
article_category = article_path.parent.name
|
156 |
with open(article_path, "r") as rf:
|
|
|
1 |
import logging
|
2 |
+
import math
|
3 |
import pathlib
|
|
|
4 |
import random
|
5 |
+
from dataclasses import dataclass
|
6 |
+
from typing import Dict, List
|
7 |
+
|
8 |
import datasets as ds
|
|
|
9 |
|
10 |
logger = logging.getLogger(__name__)
|
11 |
|
|
|
27 |
_DOWNLOAD_URL = "https://www.rondhuit.com/download/ldcc-20140209.tar.gz"
|
28 |
|
29 |
|
30 |
+
@dataclass
|
31 |
class LivedoorNewsCorpusConfig(ds.BuilderConfig):
|
32 |
+
train_ratio: float = 0.8
|
33 |
+
val_ratio: float = 0.1
|
34 |
+
test_ratio: float = 0.1
|
35 |
+
shuffle: bool = False
|
36 |
+
random_state: int = 0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
|
38 |
+
def __post_init__(self):
|
39 |
+
assert self.train_ratio + self.val_ratio + self.test_ratio == 1.0
|
40 |
|
41 |
|
42 |
class LivedoorNewsCorpusDataset(ds.GeneratorBasedBuilder):
|
43 |
+
VERSION = ds.Version("1.0.0")
|
44 |
+
BUILDER_CONFIG_CLASS = LivedoorNewsCorpusConfig
|
|
|
|
|
45 |
BUILDER_CONFIGS = [
|
46 |
LivedoorNewsCorpusConfig(
|
47 |
+
version=VERSION,
|
48 |
description="Livedoor ニュースコーパス",
|
49 |
)
|
50 |
]
|
|
|
86 |
article_paths = list(dataset_root_dir.glob("*/**/*.txt"))
|
87 |
article_paths = list(filter(lambda p: p.name != "LICENSE.txt", article_paths))
|
88 |
|
89 |
+
config: LivedoorNewsCorpusConfig = self.config
|
90 |
+
|
91 |
+
if config.shuffle:
|
92 |
+
random.seed(config.random_state)
|
93 |
random.shuffle(article_paths)
|
94 |
|
95 |
num_articles = len(article_paths)
|
96 |
+
num_tng = math.ceil(num_articles * config.train_ratio)
|
97 |
+
num_val = math.ceil(num_articles * config.val_ratio)
|
98 |
+
num_tst = math.ceil(num_articles * config.test_ratio)
|
99 |
|
100 |
tng_articles = article_paths[:num_tng]
|
101 |
val_articles = article_paths[num_tng : num_tng + num_val]
|
|
|
133 |
return example_dict
|
134 |
|
135 |
def _generate_examples(self, article_paths: List[pathlib.Path]): # type: ignore[override]
|
|
|
136 |
for i, article_path in enumerate(article_paths):
|
137 |
article_category = article_path.parent.name
|
138 |
with open(article_path, "r") as rf:
|