-
Notifications
You must be signed in to change notification settings - Fork 1
/
my_dataloader.py
158 lines (120 loc) · 5.8 KB
/
my_dataloader.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
import torch
from torchtext import data
from torchtext import datasets
import torch.nn as nn
import torch.optim as optim
import ipdb
import dill as pickle
import numpy as np
import random
def get_wiki_test_data(args, device, TEXT, LABEL):
#Vocabulary must stay consistent. fields can not be extended again
task = args.task
test_path = {"left":"test_l2.json", "mid":"test_m2.json", "right":"test_r2.json"}
test_path = test_path[args.wiki]
try:
# ipdb.set_trace()
file = open("../data/"+task+f"/dump_{args.wiki}_2.pkl","rb")
pickle_load = pickle.load(file)
fields={'s': ('text', TEXT), 'class': ('label', LABEL)}
fields, field_dict = [], fields
for field in field_dict.values():
if isinstance(field, list):
fields.extend(field)
else:
fields.append(field)
test_data = data.Dataset(pickle_load, fields=fields)
except:
file = open("../data/"+task+f"/dump_{args.wiki}_2.pkl","wb")
test_data = data.TabularDataset(
path='../data/'+task + '/' + test_path, format='json',
fields={'s': ('text', TEXT), 'class': ('label', LABEL)})
pickle.dump(list(test_data), file)
file.close()
test_iterator = data.BucketIterator(
dataset=test_data, batch_size=args.batch_size,
sort_within_batch = True, sort_key = lambda x: len(x.text),
device = device)
return test_iterator
def get_ood_test_data(args, device, TEXT, LABEL):
#Vocabulary must stay consistent. fields can not be extended again
task = args.task
test_path = {"left":"test_l2.json", "mid":"test_m2.json", "right":"test_r2.json"}
test_path = test_path[args.wiki]
# ipdb.set_trace()
file = open("../data/"+task+f"/dump_{args.wiki}.pkl","rb")
pickle_load = pickle.load(file)
# train_list, valid_list, test_list = pickle_load
fields={'s': ('text', TEXT), 'class': ('label', LABEL)}
fields, field_dict = [], fields
for field in field_dict.values():
if isinstance(field, list):
fields.extend(field)
else:
fields.append(field)
train_list, valid_list, test_list = pickle_load
test_data = data.Dataset(test_list, fields=fields)
file.close()
test_iterator = data.BucketIterator(
dataset=test_data, batch_size=args.batch_size,
sort_within_batch = True, sort_key = lambda x: len(x.text),
device = device)
return test_iterator
def load_pickle(args, TEXT, LABEL):
file = open("../data/"+args.task+f"/dump_{args.wiki}.pkl","rb") if not args.debug else open("../data/"+args.task+"/debug_dump.pkl","rb")
train_list, valid_list, test_list = pickle.load(file)
file.close()
fields={'s': ('text', TEXT), 'class': ('label', LABEL)}
fields, field_dict = [], fields
for field in field_dict.values():
if isinstance(field, list):
fields.extend(field)
else:
fields.append(field)
train_data, valid_data, test_data = data.Dataset(train_list, fields=fields), data.Dataset(valid_list, fields=fields), data.Dataset(test_list, fields=fields)
return train_data, valid_data, test_data
def dump_pickle(args, TEXT, LABEL):
train_path = {"none": "train.json", "left":"train_l.json", "mid":"train_m.json", "right":"train_r.json"}
train_path = train_path[args.wiki]
test_path = {"none": "test.json", "left":"test_l.json", "mid":"test_m.json", "right":"test_r.json"}
test_path = test_path[args.wiki]
dev_path = {"none": "dev.json", "left":"dev_l.json", "mid":"dev_m.json", "right":"dev_r.json"}
dev_path = dev_path[args.wiki]
task = args.task
print ("WARNING: Pickle Load Unsuccessful. Training time will increase")
if args.debug:
train_path = dev_path = test_path = 'small.json'
train_data, valid_data, test_data = data.TabularDataset.splits(
path='../data/'+task, train=train_path,
validation=dev_path, test=test_path, format='json',
fields={'s': ('text', TEXT), 'class': ('label', LABEL)})
file = open("../data/"+task+f"/dump_{args.wiki}.pkl","wb") if not args.debug else open("../data/"+task+"/debug_dump.pkl","wb")
train_list, valid_list, test_list = list(train_data), list(valid_data), list(test_data)
random.shuffle(train_list); random.shuffle(valid_list); random.shuffle(test_list)
pickle.dump([train_list[:25000], valid_list[:25000], test_list[:25000]], file)
file.close()
def get_data(args, MAX_VOCAB_SIZE, device):
task = args.task
print(task)
TEXT = data.Field(tokenize = 'spacy', include_lengths = True)
LABEL = data.LabelField()
try:
train_data, valid_data, test_data = load_pickle(args, TEXT, LABEL)
except:
dump_pickle(args, TEXT, LABEL)
train_data, valid_data, test_data = load_pickle(args, TEXT, LABEL)
num_examples = int(1000*float(args.data_size[:-1]))
train_data.examples = train_data.examples[:num_examples]
if not args.gradients:
valid_data.examples = valid_data.examples[:num_examples]
LABEL.build_vocab(train_data)
vec = "glove.6B.100d" if args.glove else None
unk_init = torch.Tensor.normal_ if args.glove else None
TEXT.build_vocab(train_data, max_size = MAX_VOCAB_SIZE, vectors = vec, unk_init = unk_init)
train_iterator, valid_iterator, test_iterator \
= data.BucketIterator.splits((train_data, valid_data, test_data),
batch_size = args.batch_size,
sort_within_batch = True,
sort_key = lambda x: len(x.text),
device = device)
return TEXT, LABEL, train_iterator, valid_iterator, test_iterator