-
Notifications
You must be signed in to change notification settings - Fork 7
/
preprocess.py
161 lines (116 loc) · 6.79 KB
/
preprocess.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
import argparse
import yaml
import os
import re
from transformers import AutoTokenizer
from skmultilearn.model_selection import IterativeStratification
from sklearn.preprocessing import MultiLabelBinarizer
import torch
import numpy as np
import pickle
from utils import SEEDS
from sklearn.utils import shuffle
def process_document(path, tokenizer, max_len=512):
if args.verbose == 1:
print("\n" + "*" * 50 + "\n")
document_ct = 0
big_document_ct = 0
unk_ct = 0
tokens_ct = 0
list_inputs = []
list_labels = []
list_masks = []
with open(path, "r", encoding="utf-8") as file:
for line in file:
labels = [int(x) for x in line.split()[:-2]]
text = file.readline()
text = re.sub(r"<.*?>", "", text)
text = re.sub(r"\s+", " ", text)
text = text.strip()
if args.verbose == 1:
print("Document nr: {}".format(document_ct + 1))
print("EuroVoc labels: {}".format(labels))
print("Text: {}".format(text))
inputs_ids = torch.tensor(tokenizer.encode(text))
if args.verbose == 1:
print("Input ids: {}\n".format(inputs_ids))
document_ct += 1
for token in inputs_ids[1: -1]:
if token == tokenizer.unk_token_id:
unk_ct += 1
tokens_ct += 1
if len(inputs_ids) > max_len:
big_document_ct += 1
inputs_ids = inputs_ids[:max_len]
list_inputs.append(inputs_ids)
list_labels.append(labels)
list_masks.append(torch.ones_like(inputs_ids))
print("Dataset stats - total documents: {}, big documents: {}, ratio: {:.4f}%".format(document_ct,
big_document_ct,
big_document_ct / document_ct * 100))
print(" - total tokens: {}, unk tokens: {}, ratio: {:.4f}%".format(tokens_ct,
unk_ct,
unk_ct / tokens_ct * 100))
return list_inputs, list_masks, list_labels
def save_splits(X, masks, y, directory):
for i, seed in enumerate(SEEDS):
stratifier = IterativeStratification(n_splits=2, order=2, sample_distribution_per_fold=[0.2, 0.8])
train_idx, aux_idx = next(stratifier.split(X, y))
train_X, train_mask, train_y = X[train_idx, :], masks[train_idx, :], y[train_idx, :]
assert train_X.shape[0] == train_mask.shape[0] == train_y.shape[0]
stratifier = IterativeStratification(n_splits=2, order=2, sample_distribution_per_fold=[0.5, 0.5])
dev_idx, test_idx = next(stratifier.split(X[aux_idx, :], y[aux_idx, :]))
dev_X, dev_mask, dev_y = X[dev_idx, :], masks[dev_idx, :], y[dev_idx, :]
test_X, test_mask, test_y = X[test_idx, :], masks[test_idx, :], y[test_idx, :]
assert dev_X.shape[0] == dev_mask.shape[0] == dev_y.shape[0]
assert test_X.shape[0] == test_mask.shape[0] == test_y.shape[0]
print("{} - Splitted the documents in - train: {}, dev: {}, test: {}".format(i, train_X.shape[0],
dev_X.shape[0],
test_X.shape[0]))
if not os.path.exists(os.path.join(args.data_path, directory, "split_{}".format(i))):
os.makedirs(os.path.join(args.data_path, directory, "split_{}".format(i)))
np.save(os.path.join(args.data_path, directory, "split_{}".format(i), "train_X.npy"), train_X)
np.save(os.path.join(args.data_path, directory, "split_{}".format(i), "train_mask.npy"), train_mask)
np.save(os.path.join(args.data_path, directory, "split_{}".format(i), "train_y.npy"), train_y)
np.save(os.path.join(args.data_path, directory, "split_{}".format(i), "dev_X.npy"), dev_X)
np.save(os.path.join(args.data_path, directory, "split_{}".format(i), "dev_mask.npy"), dev_mask)
np.save(os.path.join(args.data_path, directory, "split_{}".format(i), "dev_y.npy"), dev_y)
np.save(os.path.join(args.data_path, directory, "split_{}".format(i), "test_X.npy"), test_X)
np.save(os.path.join(args.data_path, directory, "split_{}".format(i), "test_mask.npy"), test_mask)
np.save(os.path.join(args.data_path, directory, "split_{}".format(i), "test_y.npy"), test_y)
X, masks, y = shuffle(X, masks, y, random_state=seed)
def process_datasets(data_path, directory, tokenizer_name):
tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)
print("Processing Acquis...")
list_ac_inputs, list_ac_masks, list_ac_labels = process_document(os.path.join(data_path, directory, "acquis.cf"),
tokenizer)
print("Processing OPOCE...")
list_op_inputs, list_op_masks, list_op_labels = process_document(os.path.join(data_path, directory, "opoce.cf"),
tokenizer)
list_inputs = list_ac_inputs + list_op_inputs
list_masks = list_ac_masks + list_op_masks
list_labels = list_ac_labels + list_op_labels
assert len(list_inputs) == len(list_masks) == len(list_labels)
mlb = MultiLabelBinarizer()
y = mlb.fit_transform(list_labels)
X = torch.nn.utils.rnn.pad_sequence(list_inputs, batch_first=True, padding_value=tokenizer.pad_token_id).numpy()
masks = torch.nn.utils.rnn.pad_sequence(list_masks, batch_first=True, padding_value=0).numpy()
with open(os.path.join(args.data_path, directory, "mlb_encoder.pickle"), "wb") as pickle_fp:
pickle.dump(mlb, pickle_fp, protocol=pickle.HIGHEST_PROTOCOL)
save_splits(X, masks, y, directory)
def preprocess_data():
with open(args.config, "r") as config_fp:
config = yaml.safe_load(config_fp)
print("Tokenizers config:\n{}\n".format(config))
for directory in os.listdir(args.data_path):
print("\nWorking on directory: {}...".format(directory))
lang = directory[:2]
print("Lang: '{}', Tokenizer: '{}'".format(lang, config[lang]))
process_datasets(args.data_path, directory, config[lang])
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--config", type=str, default="pyeurovoc/configs/models.yml", help="Tokenizer used for each language.")
parser.add_argument("--data_path", type=str, default="data/eurovoc", help="Path to the EuroVoc data.")
parser.add_argument("--verbose", type=int, default=0)
args = parser.parse_args()
preprocess_data()