-
Notifications
You must be signed in to change notification settings - Fork 0
/
concatenate_datasets_and_clean.py
226 lines (193 loc) · 7.97 KB
/
concatenate_datasets_and_clean.py
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
"""
Step1: Go inside datasets/ and execute prepare_datasets.sh (./prepare_datasets.sh) Give necessary executable permissions if needed
Step2: provide full path to dataset scripts below (/home/<user>/./...), relative path doesn't work with datasets lib.
"""
import os
import sys
import regex
from datasets import load_dataset, concatenate_datasets
# base_path = os.path.expanduser("~")
base_path = os.getcwd()
NUM_PROC = 1
DATASET_SAVE_PATH = ""
def concatenate_hindi_text_short_summarization_corpus_row(example, cols):
text = ""
for col in cols:
if example[col]:
text += example[col]
example["text"] = text
return example
def preprocess_hindi_text_short_summarization_corpus(dataset):
cols = DATASET_DICT["hindi-text-short-summarization-corpus"]["cols_to_concatenate"]
remove_cols = DATASET_DICT["hindi-text-short-summarization-corpus"]["cols_to_remove"]
dataset = dataset.map(lambda x: concatenate_hindi_text_short_summarization_corpus_row(x, cols),
remove_columns=remove_cols, num_proc=NUM_PROC)
return dataset
def concatenate_hindi_text_short_and_large_summarization_corpus(example,cols):
text = ""
for col in cols:
if example[col]:
text += example[col]
example["text"] = text
return example
def preprocess_hindi_text_short_and_large_summarization_corpus(dataset):
cols = DATASET_DICT["hindi-text-short-and-large-summarization-corpus"]["cols_to_concatenate"]
remove_cols = DATASET_DICT["hindi-text-short-and-large-summarization-corpus"]["cols_to_remove"]
dataset = dataset.map(lambda x: concatenate_hindi_text_short_and_large_summarization_corpus(x,cols),
remove_columns=remove_cols, num_proc=NUM_PROC)
return dataset
def concatenate_indic_glue_wiki_ner_row(example, col):
text = " ".join(example[col])
example["text"] = text
return example
def preprocess_indic_glue_wiki_ner(dataset):
# Only one column containing list of words eg: ["hello", "world"]
col = DATASET_DICT["indic-glue"]["cols_to_concatenate"][0]
remove_cols = DATASET_DICT["indic-glue"]["cols_to_remove"]
dataset = dataset.map(lambda x: concatenate_indic_glue_wiki_ner_row(x, col),
remove_columns=remove_cols, num_proc=NUM_PROC)
return dataset
def concatenate_samanantar_row(example, cols):
text = ""
for col in cols:
if example[col]:
text += example[col]
example["text"] = text
return example
def preprocess_samanantar(dataset):
cols = DATASET_DICT["samanantar"]["cols_to_concatenate"]
remove_cols = DATASET_DICT["samanantar"]["cols_to_remove"]
dataset = dataset.map(lambda x: concatenate_samanantar_row(x, cols),
remove_columns=remove_cols, num_proc=NUM_PROC)
return dataset
def concatenate_oldnewspapershindi_row(example, cols):
text = ""
for col in cols:
if example[col]:
text += example[col]
example["text"] = text
return example
def preprocess_oldnewspapershindi(dataset):
cols = DATASET_DICT["oldnewspapershindi"]["cols_to_concatenate"]
remove_cols = DATASET_DICT["oldnewspapershindi"]["cols_to_remove"]
dataset = dataset.map(lambda x: concatenate_oldnewspapershindi_row(x, cols),
remove_columns=remove_cols, num_proc=NUM_PROC)
return dataset
def concatenate_hindi_wiki_articles_172k_row(example, cols):
text = ""
for col in cols:
if example[col]:
text += example[col]
example["text"] = text
return example
def preprocess_hindi_wiki_articles_172k(dataset):
cols = DATASET_DICT["hindi-wikipedia-articles-172k"]["cols_to_concatenate"]
remove_cols = DATASET_DICT["hindi-wikipedia-articles-172k"]["cols_to_remove"]
dataset = dataset.map(lambda x: concatenate_hindi_wiki_articles_172k_row(x, cols),
remove_columns=remove_cols)
return dataset
def preprocess_oscar(dataset):
dataset = dataset.remove_columns(DATASET_DICT["oscar"]["cols_to_remove"])
return dataset
def replace_non_devanagiri(example):
example["text"] = regex.sub(r'\P{Devanagari}+', ' ', example["text"])
return example
# NOTE: Adjust these paths to reflect full paths appropriately
DATASET_DICT = {
"hindi-text-short-summarization-corpus": {
"is_custom": True,
"path": base_path + "/datasets/hindi-text-short-summarization-corpus",
"split_names": ["train", "test"],
"cols_to_concatenate": ["headline", "article"],
"cols_to_remove": ["headline", "article"],
"configuration": None,
"preprocess_fn": preprocess_hindi_text_short_summarization_corpus
},
"hindi-text-short-and-large-summarization-corpus": {
"is_custom": True,
"path": base_path + "/datasets/hindi-text-short-and-large-summarization-corpus",
"split_names": ["train", "test"],
"cols_to_concatenate": ["headline", "article"],
"cols_to_remove": ["headline", "article","summary"],
"configuration": None,
"preprocess_fn": preprocess_hindi_text_short_and_large_summarization_corpus
},
"indic-glue": {
"is_custom": True,
"path": base_path + "/datasets/indic-glue",
"split_names": ["train", "test"],
"configuration": "wiki-ner.hi",
"cols_to_concatenate": ["tokens"],
"cols_to_remove": ["tokens", "ner_tags", "additional_info"],
"preprocess_fn": preprocess_indic_glue_wiki_ner
},
"samanantar": {
"is_custom": True,
"path": base_path + "/datasets/samanantar",
"split_names": ["train"],
"cols_to_concatenate": ["text"],
"cols_to_remove": [],
"configuration": None,
"preprocess_fn": preprocess_samanantar
},
"oldnewspapershindi": {
"is_custom": True,
"path": base_path + "/datasets/oldnewspapershindi",
"split_names": ["train"],
"cols_to_concatenate": ["text"],
"cols_to_remove": ["source"],
"configuration": None,
"preprocess_fn": preprocess_oldnewspapershindi
},
"oscar": {
"is_custom": False,
"path": "oscar",
"split_names": ["train"],
"cols_to_concatenate": ["text"],
"cols_to_remove": ["id"],
"configuration": "unshuffled_deduplicated_hi",
"preprocess_fn": preprocess_oscar
},
"hindi-wikipedia-articles-172k": {
"is_custom": True,
"path": base_path + "/datasets/hindi-wikipedia-articles-172k",
"split_names": ["train"],
"cols_to_concatenate": ["text"],
"cols_to_remove": ["id"],
"configuration": None,
"preprocess_fn": preprocess_hindi_wiki_articles_172k
},
}
def load_and_concatenate(datasets_list, print_test_row=False):
processed_datasets = []
for dataset_id in datasets_list:
if dataset_id not in DATASET_DICT:
print("ERROR dataset config not found", dataset_id)
sys.exit(0)
for split_name in DATASET_DICT[dataset_id]["split_names"]:
dataset = load_dataset(DATASET_DICT[dataset_id]["path"],
DATASET_DICT[dataset_id]["configuration"], split=split_name)
processed_dataset = DATASET_DICT[dataset_id]["preprocess_fn"](dataset)
processed_datasets.append(processed_dataset)
if print_test_row:
print(processed_datasets)
for d in processed_datasets:
print(d[0])
concatenated_dataset = concatenate_datasets(processed_datasets)
return concatenated_dataset
datasets_list = [
"hindi-text-short-summarization-corpus",
"hindi-text-short-and-large-summarization-corpus",
"indic-glue",
"samanantar",
"oldnewspapershindi",
"oscar",
"hindi-wikipedia-articles-172k",
]
dataset = load_and_concatenate(datasets_list, print_test_row=True)
shuffle_dataset = dataset.shuffle(seed=42)
print("Removing non Devanagari characters")
shuffle_dataset = shuffle_dataset.map(replace_non_devanagiri) #, num_proc=os.cpu_count() - 1)
shuffle_dataset.save_to_disk(DATASET_SAVE_PATH)
print("Total rows:", len(shuffle_dataset))
print("Sample: ", shuffle_dataset[42]["text"])