-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_simple_baselines.py
271 lines (217 loc) · 8.45 KB
/
run_simple_baselines.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
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
from abc import ABC, abstractstaticmethod
from pathlib import Path
import numpy as np
import pandas as pd
import torch
from torch.utils.data import DataLoader, Dataset, TensorDataset
from tqdm import trange
from typing_extensions import Literal
from ethicml.algorithms.inprocess import compute_instance_weights
from ethicml.utility import DataTuple, Prediction
from nifr.configs import SharedArgs
from nifr.data import load_dataset
from nifr.models import Classifier
from nifr.models.configs.classifiers import fc_net, mp_32x32_net, mp_64x64_net
from nifr.optimisation import get_data_dim, compute_metrics
from nifr.utils import random_seed
BASELINE_METHODS = Literal["naive", "kamiran"]
class IntanceWeightedDataset(Dataset):
def __init__(self, dataset, instance_weights):
super().__init__()
if len(dataset) != len(instance_weights):
raise ValueError("Number of instance weights must equal the number of data samples.")
self.dataset = dataset
self.instance_weights = instance_weights
def __len__(self):
return len(self.dataset)
def __getitem__(self, index):
data = self.dataset[index]
if not isinstance(data, tuple):
data = (data,)
iw = self.instance_weights[index]
if not isinstance(iw, tuple):
iw = (iw,)
return data + iw
class BaselineArgs(SharedArgs):
# General data set settings
greyscale: bool = True
# Optimization settings
epochs: int = 40
test_batch_size: int = 1000
lr: float = 1e-3
# Misc settings
method: BASELINE_METHODS = "naive"
pred_s: bool = False
save_dir: str = "experiments/baseline"
def process_args(self):
if self.method == "kamiran":
if self.dataset == "cmnist":
raise ValueError(
"Kamiran & Calders reweighting scheme can only be used with binary sensitive and target attributes."
)
elif self.task_mixing_factor % 1 == 0:
raise ValueError(
"Kamiran & Calders reweighting scheme can only be used when there is at least one sample available for each sensitive/target attribute combination."
)
return super().process_args()
def get_instance_weights(dataset, batch_size):
s_all, y_all = [], []
for _, s, y in DataLoader(dataset, batch_size=batch_size):
s_all.append(s.numpy())
y_all.append(y.numpy())
s_all = np.concatenate(s_all, axis=0)
y_all = np.concatenate(y_all, axis=0)
s = pd.DataFrame(s_all, columns=["sens"])
y = pd.DataFrame(y_all, columns=["labels"])
labels = DataTuple(x=y, s=s, y=y)
instance_weights = compute_instance_weights(labels).to_numpy()
instance_weights = torch.as_tensor(instance_weights).view(-1)
instance_weights = TensorDataset(instance_weights)
return instance_weights
class Trainer(ABC):
@abstractstaticmethod
def __call__(classifier, train_loader, test_loader, epochs, device, pred_s=False):
...
class TrainNaive(Trainer):
@staticmethod
def __call__(classifier, train_loader, test_loader, epochs, device, pred_s=False):
pbar = trange(epochs)
for _ in pbar:
classifier.train()
for x, s, y in train_loader:
if pred_s:
target = s
else:
target = y
x = x.to(device)
target = target.to(device)
classifier.zero_grad()
loss, _ = classifier.routine(x, target)
loss.backward()
classifier.step()
pbar.close()
class TrainKamiran(Trainer):
@staticmethod
def __call__(classifier, train_loader, test_loader, epochs, device, pred_s=False):
pbar = trange(epochs)
for _ in pbar:
classifier.train()
for x, s, y, iw in train_loader:
if pred_s:
target = s
else:
target = y
x = x.to(device)
target = target.to(device)
iw = iw.to(device)
classifier.zero_grad()
loss, _ = classifier.routine(x, target, instance_weights=iw)
loss.backward()
classifier.step()
pbar.close()
def run_baseline(args):
use_gpu = torch.cuda.is_available() and not args.gpu < 0
random_seed(args.seed, use_gpu)
device = torch.device(f"cuda:{args.gpu}" if use_gpu else "cpu")
print(f"Running on {device}")
# Load the datasets and wrap with dataloaders
datasets = load_dataset(args)
train_data = datasets.task_train
test_data = datasets.task
if args.method == "kamiran":
instance_weights = get_instance_weights(train_data, batch_size=args.test_batch_size)
train_data = IntanceWeightedDataset(train_data, instance_weights=instance_weights)
train_loader = DataLoader(
train_data,
batch_size=args.batch_size,
pin_memory=True,
shuffle=True,
num_workers=args.num_workers,
)
test_loader = DataLoader(
test_data,
batch_size=args.test_batch_size,
pin_memory=True,
shuffle=False,
num_workers=args.num_workers,
)
input_shape = get_data_dim(train_loader)
# Construct the network
if args.dataset == "cmnist":
classifier_fn = mp_32x32_net
elif args.dataset == "adult":
def adult_fc_net(in_dim, target_dim):
encoder = fc_net(in_dim, 35, hidden_dims=[35])
classifier = torch.nn.Linear(35, target_dim)
return torch.nn.Sequential(encoder, classifier)
classifier_fn = adult_fc_net
else:
classifier_fn = mp_64x64_net
target_dim = datasets.s_dim if args.pred_s else datasets.y_dim
classifier: Classifier = Classifier(
classifier_fn(input_shape[0], target_dim),
num_classes=2 if target_dim == 1 else target_dim,
optimizer_kwargs={"lr": args.lr, "weight_decay": args.weight_decay},
)
classifier.to(device)
if args.method == "kamiran":
train_fn = TrainKamiran()
else:
train_fn = TrainNaive()
train_fn(
classifier,
train_loader=train_loader,
test_loader=test_loader,
epochs=args.epochs,
device=device,
pred_s=False,
)
preds, ground_truths, sens = classifier.predict_dataset(test_data, device=device)
preds = Prediction(pd.Series(preds))
ground_truths = DataTuple(
x=pd.DataFrame(sens, columns=["sens"]),
s=pd.DataFrame(sens.numpy().astype(np.float32), columns=["sens"]),
y=pd.DataFrame(ground_truths, columns=["labels"]),
)
full_name = f"{args.dataset}_{args.method}_baseline"
if args.dataset == "cmnist":
full_name += "_greyscale" if args.greyscale else "_color"
elif args.dataset == "celeba":
if len(args.celeba_sens_attr) > 1:
full_name += "_" + " ".join(str(v) for v in args.celeba_sens_attr)
else:
full_name += f"_{args.celeba_sens_attr[0]}"
full_name += f"_{args.celeba_target_attr}"
full_name += f"_{args.epochs}epochs.csv"
metrics = compute_metrics(args, preds, ground_truths, "baselines", 0, use_wandb=False)
print(f"Results for {full_name}:")
print("\n".join(f"\t\t{key}: {value:.4f}" for key, value in metrics.items()))
print()
if args.save is not None:
save_to_csv = Path(args.save_dir)
save_to_csv.mkdir(exist_ok=True)
assert isinstance(save_to_csv, Path)
results_path = save_to_csv / full_name
if args.dataset == "cmnist":
value_list = ",".join([str(args.scale)] + [str(v) for v in metrics.values()])
else:
value_list = ",".join(
[str(args.task_mixing_factor)] + [str(v) for v in metrics.values()]
)
if results_path.is_file():
with results_path.open("a") as f:
f.write(value_list + "\n")
else:
with results_path.open("w") as f:
if args.dataset == "cmnist":
f.write(",".join(["Scale"] + [str(k) for k in metrics.keys()]) + "\n")
else:
f.write(",".join(["Mix_fact"] + [str(k) for k in metrics.keys()]) + "\n")
f.write(value_list + "\n")
def main():
args = BaselineArgs(explicit_bool=True, underscores_to_dashes=True)
args.parse_args()
print(args)
run_baseline(args=args)
if __name__ == "__main__":
main()