File size: 9,205 Bytes
54fa0c8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
import argparse
import json
import logging
import random
import time
import numpy as np
from functools import partial
from pprint import pformat

from datasets import load_dataset
from datasets.utils.logging import set_verbosity_info

from manual_sharding import save_manual_shards
from utils import get_replacements, redact_pii_batch


REPONAME_TOKEN = "<reponame>"
FILENAME_TOKEN = "<filename>"
STARS_TOKEN = "<gh_stars>"


def get_num_stars_bucket(num_stars: int) -> str:
    if num_stars is None or num_stars == 0:
        return "0"
    elif num_stars <= 10:
        return "1-10"
    elif num_stars <= 100:
        return "10-100"
    elif num_stars <= 1000:
        return "100-1000"
    else:
        return "1000+"


def content_with_meta(example):
    res = ""
    if np.random.binomial(n=1, p=0.2):
        res += f"{REPONAME_TOKEN}{example['max_stars_repo_name']}"
    if np.random.binomial(n=1, p=0.2):
        res += f"{FILENAME_TOKEN}{example['max_stars_repo_path']}"
    if np.random.binomial(n=1, p=0.2):
        num_stars = get_num_stars_bucket(example["max_stars_count"])
        res += f"{STARS_TOKEN}{num_stars}"
    if len(res) > 0:
        res += "\n"
    res += example["content"]

    return {"content_with_meta": res}


def parseArgs():
    parser = argparse.ArgumentParser(description="PII detection and redaction")
    parser.add_argument(
        "--dataset_name",
        default="bigcode/pii-for-code",
        type=str,
        help="HF repo name/path of the dataset.",
    )
    # add arg true add metadata
    parser.add_argument(
        "--add_metadata",
        action="store_true",
        help="If set, we add metadata to the text",
    )
    parser.add_argument(
        "--num_load_proc",
        default=64,
        type=int,
        help="Number of processes to use for loading the dataset",
    )
    parser.add_argument(
        "--text_column",
        default="content",
        type=str,
        help="Text column to use, if will be renamed to content",
    )
    parser.add_argument(
        "--split",
        default="train",
        type=str,
        help="Dataset split to process",
    )
    parser.add_argument(
        "--batch_size",
        default=100,
        type=int,
        help="Batch size for the PII detection/redaction",
    )
    parser.add_argument(
        "--seed",
        default=0,
        type=int,
        help="Seed for random",
    )
    parser.add_argument(
        "--num_proc",
        default=96,
        type=int,
        help="Number of processes to use for the PII detection/redaction",
    )
    parser.add_argument(
        "--no_redaction",
        action="store_true",
        help="If set, we don't perform redaction",
    )
    parser.add_argument(
        "--load_replacements",
        default=True,
        help="If set, we load the replacements from file replacements.json",
    )
    parser.add_argument(
        "--add_reference_text",
        default=True,
        type=bool,
        help="If True we add the reference text with PII between delimiters \
        in the redacted text -used for visualization-",
    )
    parser.add_argument(
        "--check_all_files",
        action="store_true",
        help="If set, we check all files, not only the ones that contain PII",
    )
    parser.add_argument(
        "--check_sampling_size",
        default=0,
        type=int,
        help="Number of samples to check for PII",
    )
    # for saving the dataset: either push to HF or save locally with datasets or save manual shards
    parser.add_argument(
        "--save_mode",
        default="manual_shards",
        type=str,
        choices=["hub", "local", "manual_shards"],
        help="How to save the dataset",
    )
    parser.add_argument(
        "--save_mode_checks",
        default="hub",
        type=str,
        choices=["hub", "local", "manual_shards"],
        help="How to save the  checks dataset",
    )
    # add argument for name of dataset on the hub
    parser.add_argument(
        "--target_dataset",
        default="bigcode-pii2",
        type=str,
        help="HF repo name of the target dataset in save_mode=hub.",
    )
    parser.add_argument(
        "--hub_username",
        default="loubnabnl",
        type=str,
        help="Username for the hub",
    )
    parser.add_argument(
        "--save_path_disk",
        default="/fsx/loubna/data/the-stack-march-no-pii",
        type=str,
        help="Path to save the dataset on disk in save_mode=local.",
    )
    return parser.parse_args()


def get_check_ds(ds, args):
    if not args.check_all_files:
        ds_checks = ds.filter(
            lambda exs: exs["modified"],
            batched=True,
            batch_size=args.batch_size,
            num_proc=args.num_proc,
        )
    else:
        ds_checks = ds
    if not args.check_sampling_size:
        sampling_size = len(ds_checks)
    idx_samples = random.sample(
        range(len(ds_checks)), min(len(ds_checks), sampling_size)
    )
    ds_checks = ds_checks.select(idx_samples)

    return ds_checks


def main():
    set_verbosity_info()
    args = parseArgs()
    logger = logging.getLogger(__name__)
    logger.setLevel(logging.INFO)
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        level=logging.INFO,
        handlers=[
            logging.FileHandler(f"logs/pii-{args.dataset_name.split('/')[-1]}.log"),
            logging.StreamHandler(),
        ],
    )
    logger.info(
        f"** The job is running with the following arguments: **\n{args}\n **** "
    )

    logger.info(f" ===== Loading {args.dataset_name} =====")
    ds = load_dataset(
        args.dataset_name,
        split=args.split,
        use_auth_token=True,
        num_proc=args.num_load_proc,
    )
    if args.text_column != "content":
        ds = ds.rename_column(args.text_column, "content")

    # redact PII in the dataset
    logger.info(f" ===== Applying PII redaction =====")
    random.seed(args.seed)

    replacements = get_replacements()
    with open("replacements.json", "w") as f:
        json.dump(replacements, f)
    logging.info(f"Using the following replacements:\n{pformat(replacements)}")
    ds = ds.map(
        partial(
            redact_pii_batch,
            replacements=replacements,
            add_references=args.add_reference_text,
        ),
        batched=True,
        batch_size=args.batch_size,
        num_proc=args.num_proc,
    )
    logging.info(f"Dataset info after PII redaction:\n{ds}")

    # check the dataset
    logger.info(
        f" ===== Checking {args.check_sampling_size} samples from those modified in the dataset ====="
    )
    ds_checks = get_check_ds(ds, args)

    # save checks dataset
    if len(ds_checks) == 0:
        logger.info("Dataset was empty. Not saving anything.")
    else:
        logger.info(f"Checks dataset info {ds_checks}")
        if args.save_mode_checks == "hub":
            logger.info(
                f"Pushing the checks dataset to the Hub as {args.target_dataset}_checks"
            )
            ds_checks.push_to_hub(args.target_dataset + "_checks", private=True)

        elif args.save_mode_checks == "local":
            logger.info(f"Saving the checks dataset to disk")
            ds_checks.save_to_disk(args.save_path_disk + "_checks")

        elif args.save_mode_checks == "manual_shards":
            logger.info(f"Saving the checks dataset in manual shards")
            save_manual_shards(
                ds_checks,
                user=args.hub_username,
                remote_dataset_repo=args.target_dataset + "_checks",
                local_dir="/fsx/loubna/data/the-stack-march-no-pii_checks",
            )

    logger.info("Removing columns that are not needed for the final dataset")
    columns = ["content", "modified", "entities"]
    if args.add_reference_text:
        columns.append("references")
    ds = ds.remove_columns(columns)
    ds = ds.rename_column("new_content", "content")
    logger.info(f"Dataset info after removing columns:\n{ds}")

    if args.add_metadata:
        logger.info(f" ===== Adding metadata =====")
        ds = ds.map(
            content_with_meta, remove_columns=["content"], num_proc=args.num_proc
        )
        ds = ds.rename_column("content_with_meta", "content")

    # save the final dataset
    if args.save_mode == "hub":
        logger.info(
            f" ===== Pushing the dataset to the Hub as: {args.target_dataset} ====="
        )
        ds.push_to_hub(args.target_dataset, private=True)

    elif args.save_mode == "local":
        logger.info(f" ===== Saving the dataset to disk =====")
        ds.save_to_disk(args.save_path_disk)

    elif args.save_mode == "manual_shards":
        logger.info(
            f" ===== Saving the dataset in manual shards to {args.save_path_disk} ====="
        )
        save_manual_shards(
            ds,
            user=args.hub_username,
            remote_dataset_repo="the-stack-no-pii-march",
            local_dir=args.save_path_disk,
        )

    logger.info(f" ===== Dataset saved successfully =====")


if __name__ == "__main__":
    main()