-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdomainbed_dataset.py
192 lines (158 loc) · 7.37 KB
/
domainbed_dataset.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
import argparse
import bisect
import hydra
from omegaconf import OmegaConf
from libs.DomainBed.domainbed.datasets import OfficeHome, DomainNet, VLCS, PACS, TerraIncognita
from libs.DomainBed.domainbed.lib import misc
from libs.DomainBed.domainbed.lib.fast_data_loader import InfiniteDataLoader, FastDataLoader
from torch.utils.data import DataLoader, Dataset
class MultiDomainDataset(Dataset):
def __init__(self, datasets, env_idxs=None):
self.datasets = datasets
self.lengths = [len(d) for d in self.datasets]
if env_idxs is None:
self.env_idxs = [i for i, d in enumerate(self.datasets) for _ in range(len(d))]
else:
self.env_idxs = env_idxs
def find_dataset(self, index, lengths):
# Create a list of cumulative sums of lengths
cumulative_lengths = [sum(lengths[:i+1]) for i in range(len(lengths))]
# Use bisect to find the position where index would fit in the cumulative list
dataset_index = bisect.bisect(cumulative_lengths, index)
# Adjust for Python's 0-based indexing
return dataset_index if dataset_index < len(lengths) else -1
def __len__(self):
return sum(len(d) for d in self.datasets)
def __getitem__(self, i):
dataset_idx = self.find_dataset(i, self.lengths)
data_idx = i - sum(self.lengths[:dataset_idx])
x = self.datasets[dataset_idx][data_idx][0]
y = self.datasets[dataset_idx][data_idx][1]
e = self.env_idxs[dataset_idx]
return x, y, e
# from libs.DomainBed.domainbed.datasets import split_dataset
def get_dataset(cfg):
name = cfg.dataset.name
if name == "officehome":
C = OfficeHome
elif name == "domainnet":
C = DomainNet
elif name == "vlcs":
C = VLCS
elif name == "pacs":
C = PACS
elif name == "terraincognita":
C = TerraIncognita
else:
raise ValueError("Unknown dataset %s" % name)
# hparams = {}
# hparams["holdout_fraction"] = 0.2
# hparams["data_augmentation"] = True
# hparams["seed"] = 1
# hparams["target_envs"] = [0]
# hparams["uda_holdout_fraction"] = 0
# hparams["class_balanced"] = False
# hparams["batch_size"] = 64
dataset_config_dict = OmegaConf.to_container(cfg.dataset, resolve=True)
dataset_config_dict["seed"] = cfg.dataset.seed
dataset = C(root=cfg.server.data_root_folder,
test_envs=cfg.dataset.target_envs,
hparams=dataset_config_dict)
in_splits = []
out_splits = []
uda_splits = []
for env_i, env in enumerate(dataset):
uda = []
out, in_ = misc.split_dataset(env,
int(len(env)*cfg.dataset.holdout_fraction),
misc.seed_hash(cfg.dataset.seed, env_i))
if env_i in cfg.dataset.target_envs:
uda, in_ = misc.split_dataset(in_,
int(len(in_)*cfg.dataset.uda_holdout_fraction),
misc.seed_hash(cfg.dataset.seed, env_i))
if cfg.dataset.class_balanced:
in_weights = misc.make_weights_for_balanced_classes(in_)
out_weights = misc.make_weights_for_balanced_classes(out)
if uda is not None:
uda_weights = misc.make_weights_for_balanced_classes(uda)
else:
in_weights, out_weights, uda_weights = None, None, None
in_splits.append((in_, in_weights))
out_splits.append((out, out_weights))
if len(uda):
uda_splits.append((uda, uda_weights))
test_splits = []
if cfg.mode == "source_pretrain":
assert cfg.indomain_test is False
if cfg.indomain_test:
# logger.info("!!! In-domain test mode On !!!")
# assert hparams["val_augment"] is False, (
# "indomain_test split the val set into val/test sets. "
# "Therefore, the val set should be not augmented."
# )
val_splits = []
for env_i, (out_split, _weights) in enumerate(out_splits):
n = len(out_split) // 2
seed = misc.seed_hash(cfg.dataset.seed, env_i)
val_split, test_split = misc.split_dataset(out_split, n, seed=misc.seed_hash(cfg.dataset.seed, env_i))
val_splits.append((val_split, None))
test_splits.append((test_split, None))
# logger.info(
# "env %d: out (#%d) -> val (#%d) / test (#%d)"
# % (env_i, len(out_split), len(val_split), len(test_split))
# )
out_splits = val_splits
in_splits = [s[0] for s in in_splits]
out_splits = [s[0] for s in out_splits]
test_splits = [s[0] for s in test_splits]
uda_splits = [s[0] for s in uda_splits]
target_envs = cfg.dataset.target_envs
source_envs = [env_i for env_i, env in enumerate(dataset) if env_i not in target_envs]
print("source_envs", source_envs)
print("target_envs", target_envs)
source_train_datasets = [env for env_i, env in enumerate(in_splits) if env_i not in target_envs]
source_val_datasets = [env for env_i, env in enumerate(out_splits) if env_i not in target_envs]
source_test_datasets = [env for env_i, env in enumerate(test_splits) if env_i not in target_envs]
target_train_datasets = [env for env_i, env in enumerate(in_splits) if env_i in target_envs]
target_val_datasets = [env for env_i, env in enumerate(out_splits) if env_i in target_envs]
target_test_datasets = [env for env_i, env in enumerate(test_splits) if env_i in target_envs]
source_train_dataset = MultiDomainDataset(source_train_datasets, source_envs)
source_val_dataset = MultiDomainDataset(source_val_datasets, source_envs)
target_train_dataset = MultiDomainDataset(target_train_datasets, target_envs)
target_val_dataset = MultiDomainDataset(target_val_datasets, target_envs)
source_test_dataset = MultiDomainDataset(source_test_datasets, source_envs)
target_test_dataset = MultiDomainDataset(target_test_datasets, target_envs)
dataset_dict = {
"source_train_dataset": source_train_dataset,
"source_val_dataset": source_val_dataset,
"target_train_dataset": target_train_dataset,
"target_val_dataset": target_val_dataset,
"source_test_dataset": source_test_dataset,
"target_test_dataset": target_test_dataset,
}
return dataset_dict
@hydra.main(config_path="configs", config_name="default", version_base="1.3.0")
def test(cfg):
dataset_dict = get_dataset(cfg)
print(dataset_dict.keys())
target_envs = cfg.dataset.target_envs
print("target_envs", target_envs)
for k, v in dataset_dict.items():
print(k, len(v))
# for split in dataset_dict:
# print(dataset_dict[split][0][0].shape, dataset_dict[split][0][1], dataset_dict[split].env_idxs)
# for i in range(len(dataset_dict[split])):
# print(dataset_dict[split][i][0].shape, dataset_dict[split][i][1], dataset_dict[split][i][2])
# print("====================================")
if __name__ == "__main__":
test()
# hparams = {}
# hparams["holdout_fraction"] = 0.2
# hparams["data_augmentation"] = True
# hparams["seed"] = 1
# hparams["target_envs"] = [0]
# hparams["uda_holdout_fraction"] = 0
# hparams["class_balanced"] = False
# hparams["batch_size"] = 64
# args = argparse.Namespace(**hparams)
# dataset = get_dataset(hparams)