ruggsea commited on
Commit
c5c2c0e
1 Parent(s): 5ad75a6

Adding dataset card

Browse files
Files changed (3) hide show
  1. README.md +81 -24
  2. UsenetArchiveIT.py +0 -188
  3. it.comp.os.win.nt_NO_VIRUS.jsonl +0 -3
README.md CHANGED
@@ -1,27 +1,84 @@
1
  ---
2
  dataset_info:
3
- features:
4
- - name: title
5
- dtype: string
6
- - name: author
7
- dtype: string
8
- - name: id
9
- dtype: int32
10
- - name: timestamp
11
- dtype: string
12
- - name: progressive_number
13
- dtype: int32
14
- - name: original_url
15
- dtype: string
16
- - name: newsgroup
17
- dtype: string
18
- - name: text
19
- dtype: string
20
- splits:
21
- - name: train
22
- path: "parquet/*.parquet"
23
- num_bytes: 72373684017
24
- num_examples: 85010057
25
- download_size: 0
26
- dataset_size: 72373684017
27
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  dataset_info:
3
+ features:
4
+ - name: title
5
+ dtype: string
6
+ - name: author
7
+ dtype: string
8
+ - name: id
9
+ dtype: int32
10
+ - name: timestamp
11
+ dtype: string
12
+ - name: progressive_number
13
+ dtype: int32
14
+ - name: original_url
15
+ dtype: string
16
+ - name: newsgroup
17
+ dtype: string
18
+ - name: text
19
+ dtype: string
20
+ splits:
21
+ - name: trainv
22
+ path: "parquet/*.parquet"
23
+ num_bytes: 72373684017
24
+ num_examples: 85010057
25
+ download_size: 0
26
+ dataset_size: 72373684017
27
  ---
28
+ # Usenet Archive IT Dataset 🇮🇹
29
+
30
+ ## Description
31
+
32
+ ### Dataset Content
33
+
34
+ This dataset contains Usenet posts from Italian language newsgroups belonging to the `it`, `it-alt` and `italia` hierarchies. The data has been archived and converted to the Parquet format for easy processing. The only preprocessing conducted on the text was the removal of source code of some malicious scripts that were present in the original data and were causing HF to flag the dataset as malicious.
35
+
36
+ This dataset contributes to the [mii-community](https://huggingface.co/mii-community) project, aimed at advancing the creation of Italian open-source Language Models (LLMs).🇮🇹 🤖
37
+
38
+ ### Descriptive Statistics
39
+
40
+ This dataset contains 85010057 posts from 11956999 threads in 539 newsgroups. Threads appear to have around 7 posts on average, with a median of 3 posts.
41
+ The posts were created between 1995 and 2024. The text of all the posts together sum up to a total of 55885335313 characters, or approximately 10-20B tokens. The average length of the posts is 657 characters, and the median length is 380 characters.
42
+
43
+ ### Languages
44
+
45
+ The dataset should contain only Italian language posts, but it is possible that some posts are in other languages. The dataset has not been language filtered, as post were expected to be in Italian.
46
+
47
+ ## Dataset Structure
48
+
49
+ Each record in the dataset has the following fields:
50
+
51
+ - `title`: The title of the post.
52
+ - `author`: The username of the author of the post.
53
+ - `id`: The unique identifier of the post.
54
+ - `timestamp`: The timestamp of the post.
55
+ - `progressive_number`: An integer identifying the thread number in the newsgroup.
56
+ - `original_url`: The URL of the original post on Google Groups.
57
+ - `newsgroup`: The name of the newsgroup the post belongs to.
58
+ - `text`: The text content of the post.
59
+
60
+ This repo contains the dataset in the Parquet format. The dataset is split into multiple Parquet files inside the `parquet` folder, each containing a portion of the records. The files are named `usenet_converted_*.parquet`, where `*` is a number indicating the order of the file.
61
+ The original jsonl lines of the data are included as well as compressed bz2 files.
62
+
63
+
64
+ ## Additional Information
65
+
66
+ ### Dataset Curators
67
+
68
+ This dataset was compiled and curated by Hugging Face users [manalog](https://huggingface.co/manalog) and [ruggsea](https://huggingface.co/ruggsea), as part of the [mii-community](https://huggingface.co/mii-community) dataset creation effort.
69
+
70
+ ### Dataset rationale
71
+
72
+ The dataset was created as part of a bigger effort to create various high-quality datasets of native Italian text, with the aim of aiding the development of Italian open-source LLMs.
73
+
74
+ The dataset is expected to be used for training and fine-tuning language models, as well as for other NLP tasks such as text classification, summarization, and translation. The column `text` contains the raw text of the posts, and the column `newsgroup` contains the name of the newsgroup the post belongs to, which can be used for classification tasks.
75
+
76
+ ## Usage
77
+
78
+ You can load the dataset directly from datasets using the `load_dataset` function. Here's an example:
79
+
80
+ ```python
81
+ from datasets import load_dataset
82
+
83
+ dataset = load_dataset("manalog/UsenetArchiveIT")
84
+ ```
UsenetArchiveIT.py DELETED
@@ -1,188 +0,0 @@
1
- from datasets import DatasetBuilder, SplitGenerator, Split, Features, Value, ClassLabel, BuilderConfig, Version, DatasetInfo, DownloadManager, ArrowBasedBuilder
2
- import glob
3
- import json
4
- import multiprocessing as mp
5
- import os
6
- import pyarrow as pa
7
- import pyarrow.parquet as pq
8
- import pandas as pd
9
- import pyarrow as pa
10
- import pyarrow.json
11
- # jsonl
12
-
13
- pattern="*.bz2"
14
-
15
- paths=glob.glob(pattern)
16
-
17
- # exclude txt files
18
-
19
- paths=[file for file in paths if not ".txt." in file]
20
-
21
- n_files=len(paths)
22
-
23
- # labels are file names without the extension .jsonl.bz2
24
-
25
- labels=[file.replace(".jsonl.bz2","") for file in paths]
26
-
27
-
28
-
29
- ## handle parquet conversion
30
-
31
- # create parquet directory
32
-
33
- dl_manager = DownloadManager()
34
-
35
- parquet_dir="parquet"
36
-
37
-
38
-
39
-
40
- def convert_jsonl_to_parquet(file_list, parquet_dir, chunk_size=100000):
41
- """Converts JSONL files to Parquet with memory efficiency.
42
-
43
- Args:
44
- file_list (list): List of JSONL file paths.
45
- parquet_dir (str): Path to store output Parquet files.
46
- chunk_size (int): Number of records to write to each Parquet file.
47
- """
48
-
49
- os.makedirs(parquet_dir, exist_ok=True) # Create output directory
50
-
51
- parquet_file_index = 0
52
- current_records = []
53
- file_index = 0
54
- for file in file_list:
55
- # try:
56
- reader = pa.json.read_json(file) # PyArrow JSON reader
57
-
58
- for batch in reader:
59
- pandas_df = batch.to_pandas()
60
- print(pandas_df.shape)
61
- current_records.extend(pandas_df.to_dict('list'))
62
- if len(current_records) >= chunk_size:
63
- table = pa.Table.from_pandas(pd.DataFrame(current_records))
64
- parquet_filename = f"output_{parquet_file_index}.parquet"
65
- parquet_path = os.path.join(parquet_dir, parquet_filename)
66
- pq.write_table(table, parquet_path)
67
-
68
- current_records = []
69
- parquet_file_index += 1
70
- # except Exception as e:
71
- # print(f"Error in file {file} with error {e}")
72
- file_index += 1
73
- print(f"Finished processing file {file_index} of {len(file_list)}")
74
- print(f"Writing last chunk to parquet file {parquet_file_index}")
75
- # Write any remaining data in the last chunk
76
- if current_records:
77
- table = pa.Table.from_pandas(pd.DataFrame(current_records))
78
- parquet_filename = f"output_{parquet_file_index}.parquet"
79
- parquet_path = os.path.join(parquet_dir, parquet_filename)
80
- pq.write_table(table, parquet_path)
81
-
82
- print(f"Conversion complete, wrote {parquet_file_index + 1} Parquet files.")
83
-
84
-
85
-
86
-
87
-
88
- class UsenetConfig(BuilderConfig):
89
- def __init__(self, version, **kwargs):
90
- ().__init__(version, **kwargs)
91
-
92
-
93
-
94
-
95
-
96
-
97
-
98
-
99
-
100
- class UsenetArchiveIt(ArrowBasedBuilder):
101
- VERSION = "1.0.0" # Example version
102
-
103
- BUILDER_CONFIG_CLASS = UsenetConfig
104
-
105
- BUILDER_CONFIGS = [
106
- UsenetConfig(
107
- name="usenet_archive_it",
108
- version=Version("1.0.0"),
109
- description="Usenet Archive-It dataset",
110
- ),
111
- ]
112
-
113
- def _info(self):
114
- # Specify dataset features here
115
- return DatasetInfo(
116
- features=Features({
117
- "title": Value("string"),
118
- "author": Value("string"),
119
- "id": Value("int32"),
120
- "timestamp": Value("string"),
121
- "progressive_number": Value("int32"),
122
- "original_url": Value("string"),
123
- "newsgroup": Value("string"), # this could be a label but difficult to get all possible labels
124
- "text": Value("string"),
125
- }),)
126
-
127
- def _split_generators(self, dl_manager):
128
- n = mp.cpu_count()//10 # Number of paths to process at a time
129
- print(f"Extracting {n} files at a time")
130
- if not os.path.isdir('parquet'):
131
- extracted_files = []
132
- for i in range(0, len(paths), n):
133
-
134
- files = paths[i:i+n]
135
- extracted_files.extend(dl_manager.extract(files, num_proc=len(files)))
136
- print(f"Extracted {files}")
137
- else:
138
- extracted_files = glob.glob(parquet_dir + "/*.parquet")
139
-
140
- return [
141
- SplitGenerator(
142
- name=Split.TRAIN,
143
- gen_kwargs={"filepath": extracted_files},
144
- ),
145
-
146
- ]
147
-
148
- def _generate_tables(self, filepath):
149
-
150
- # print("Filepath: ", filepath)
151
-
152
- # if parquet files are not present, convert jsonl to parquet
153
- if not os.path.exists(parquet_dir):
154
- print("Generating parquet files from jsonl files...")
155
- convert_jsonl_to_parquet(filepath, parquet_dir)
156
-
157
- # read parquet files
158
- parquet_files=glob.glob(parquet_dir+"/*.parquet")
159
-
160
-
161
- for index, file in enumerate(parquet_files):
162
- table = pq.read_table(file)
163
- yield index, table
164
-
165
-
166
- # for file in parquet_files:
167
- # table = pq.read_table(file)
168
- # df = table.to_pandas()
169
- # for index, row in df.iterrows():
170
- # yield index, row.to_dict()
171
-
172
-
173
- # Yields (key, example) tuples from the dataset
174
- # id=0
175
- # for file in filepath:
176
- # # Open and yield examples from the compressed JSON files
177
- # with open(file, "r") as f:
178
- # for i, line in enumerate(f):
179
- # try:
180
- # data = json.loads(line)
181
- # yield id, data
182
- # id+=1
183
- # except Exception as e:
184
- # print(f"Error in file {file} at line {i} with error {e}")
185
-
186
-
187
- # Finally, set the name of the dataset to match the script name
188
- datasets = UsenetArchiveIt
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
it.comp.os.win.nt_NO_VIRUS.jsonl DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:86763db842a8d053b2706af57259a99e23ef2191c7499dad9749a0c2888ddb16
3
- size 5569638