-
Notifications
You must be signed in to change notification settings - Fork 56
/
data_loader.py
146 lines (116 loc) · 4.34 KB
/
data_loader.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
import json
import numpy as np
from collections import Counter
from torch.utils.data import Dataset
import pandas as pd
from tqdm import tqdm
from . import utils
import torch
def get_sample_weights(labels):
counter = Counter(labels)
counter = dict(counter)
for k in counter:
counter[k] = 1 / counter[k]
sample_weights = np.array([counter[l] for l in labels])
return sample_weights
def load_data(args):
# chunk your dataframes in small portions
chunks = pd.read_csv(
args.data_path,
usecols=[args.text_column, args.label_column],
chunksize=args.chunksize,
encoding=args.encoding,
nrows=args.max_rows,
sep=args.sep,
)
texts = []
labels = []
for df_chunk in tqdm(chunks):
aux_df = df_chunk.copy()
aux_df = aux_df.sample(frac=1)
aux_df = aux_df[~aux_df[args.text_column].isnull()]
aux_df = aux_df[(aux_df[args.text_column].map(len) > 1)]
aux_df["processed_text"] = aux_df[args.text_column].map(
lambda text: utils.process_text(args.steps, text)
)
texts += aux_df["processed_text"].tolist()
labels += aux_df[args.label_column].tolist()
if bool(args.group_labels):
if bool(args.ignore_center):
label_ignored = args.label_ignored
clean_data = [
(text, label)
for (text, label) in zip(texts, labels)
if label not in [label_ignored]
]
texts = [text for (text, label) in clean_data]
labels = [label for (text, label) in clean_data]
labels = list(map(lambda l: {1: 0, 2: 0, 4: 1, 5: 1}[l], labels))
else:
labels = list(map(lambda l: {1: 0, 2: 0, 3: 1, 4: 2, 5: 2}[l], labels))
if bool(args.balance):
counter = Counter(labels)
keys = list(counter.keys())
values = list(counter.values())
count_minority = np.min(values)
balanced_labels = []
balanced_texts = []
for key in keys:
balanced_texts += [
text for text, label in zip(texts, labels) if label == key
][: int(args.ratio * count_minority)]
balanced_labels += [
label for text, label in zip(texts, labels) if label == key
][: int(args.ratio * count_minority)]
texts = balanced_texts
labels = balanced_labels
number_of_classes = len(set(labels))
print(
f"data loaded successfully with {len(texts)} rows and {number_of_classes} labels"
)
print("Distribution of the classes", Counter(labels))
sample_weights = get_sample_weights(labels)
return texts, labels, number_of_classes, sample_weights
class MyDataset(Dataset):
def __init__(self, texts, labels, args):
self.texts = texts
self.labels = labels
self.length = len(self.texts)
self.vocabulary = args.alphabet + args.extra_characters
self.number_of_characters = args.number_of_characters + len(
args.extra_characters
)
self.max_length = args.max_length
self.preprocessing_steps = args.steps
self.identity_mat = np.identity(self.number_of_characters)
def __len__(self):
return self.length
def __getitem__(self, index):
raw_text = self.texts[index]
data = np.array(
[
self.identity_mat[self.vocabulary.index(i)]
for i in list(raw_text)[::-1]
if i in self.vocabulary
],
dtype=np.float32,
)
if len(data) > self.max_length:
data = data[: self.max_length]
elif 0 < len(data) < self.max_length:
data = np.concatenate(
(
data,
np.zeros(
(self.max_length - len(data), self.number_of_characters),
dtype=np.float32,
),
)
)
elif len(data) == 0:
data = np.zeros(
(self.max_length, self.number_of_characters), dtype=np.float32
)
label = self.labels[index]
data = torch.Tensor(data)
return data, label