-
Notifications
You must be signed in to change notification settings - Fork 0
/
local_grid.py
207 lines (178 loc) · 7.09 KB
/
local_grid.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
import argparse
from dataclasses import dataclass, asdict
from typing import Dict, List, Any, Union
from training import train_eval, TrainConfig
from models import GraphMLP, GCN
from data import (
generate_dataloaders,
get_dataset,
get_test_val_train_split,
get_test_val_train_mask,
)
from utils import get_device, parse_callable_string, BasicLogger, CSVLogger
import torchmetrics
from torch.nn.functional import cross_entropy
import torch
import numpy as np
def main():
parser = argparse.ArgumentParser()
parser.add_argument("-n", "--name", required=True)
parser.add_argument("-e", "--epochs", default=100, type=int)
parser.add_argument("-m", "--model", required=True)
parser.add_argument("--batch_size", default=128, type=int)
parser.add_argument("--dataset", required=True)
parser.add_argument("-s", "--split", default=None, type=int)
parser.add_argument(
"--control_type", default="null", type=str, choices=["null", "gcn", "mp"]
)
parser.add_argument(
"--control_edges", default="adj", type=str, choices=["adj", "dense_subset"]
)
parser.add_argument(
"--control_metric",
default="b_centrality",
type=str,
choices=["degree", "b_centrality", "pr_centrality", "curvature"],
)
parser.add_argument("--control_k", default="1", type=str)
parser.add_argument("--control_self_adj", action="store_true")
parser.add_argument("-t", "--time_inv", action="store_true", default=False)
parser.add_argument("-l", "--linear", action="store_true", default=False)
parser.add_argument("--hidden_dim", default=128, type=int)
parser.add_argument("--conv_depth", default=2, type=int)
parser.add_argument(
"--norm", default="layernorm", choices=[None, "batchnorm", "layernorm"]
)
parser.add_argument("--save_models", action="store_true")
parser.add_argument("--control_init", default=None, type=float)
args = parser.parse_args()
dataset, is_node_classifier = get_dataset(
args.dataset,
args.control_type,
args.control_edges,
args.control_metric,
parse_callable_string(args.control_k),
args.control_self_adj,
)
if args.norm == "batchnorm":
norm = lambda channels: torch.nn.BatchNorm1d(momentum=args.bn_momentum)
elif args.norm == "layernorm":
norm = torch.nn.LayerNorm
elif args.norm is None:
norm = None
else:
raise ValueError("Norm must be None, layernorm or batchnorm")
if args.model == "mlp":
model_factory = lambda dropout_rate: GraphMLP(
input_dim=dataset[0].x.shape[1],
output_dim=dataset.num_classes,
hidden_dim=args.hidden_dim,
dropout_rate=dropout_rate,
is_node_classifier=is_node_classifier,
norm=norm,
)
elif args.model == "gcn":
model_factory = lambda dropout_rate: GCN(
input_dim=dataset[0].x.shape[1],
output_dim=dataset.num_classes,
hidden_dim=args.hidden_dim,
conv_depth=args.conv_depth,
dropout_rate=dropout_rate, # passed during hyperparameter tuning
linear=args.linear,
time_inv=args.time_inv,
control_type=args.control_type,
is_node_classifier=is_node_classifier,
norm=norm,
control_init=args.control_init,
)
else:
raise ValueError(f"Model name {args.model} not recognized")
device = get_device()
accuracy_function = torchmetrics.Accuracy(
"multiclass", num_classes=dataset.num_classes
).to(device)
if args.split is None:
splits = range(5)
else:
splits = args.split
run_logger = CSVLogger(args.name+'_run.csv', [
"lr",
"weight_decay",
"dropout_rate",
"split",
"best_val_loss",
"test_loss",
"test_metric",
])
epoch_logger = CSVLogger(args.name+'_epoch.csv', [
"lr",
"weight_decay",
"dropout_rate",
"split",
"epoch",
"train_loss",
"val_loss",
"train_metric",
"val_metric"])
for lr in [1e-4, 1e-3, 1e-2]:
for weight_decay in [0.0, 1e-6, 1e-5]:
for dropout_rate in [0.0, 0.2, 0.4]:
mean_stats = {"best_val_loss": [],
"test_loss": [],
"test_metric": []}
for split in splits:
log_const = {
"lr": lr,
"weight_decay": weight_decay,
"dropout_rate": dropout_rate,
"split": split
}
epoch_logger.update_const(log_const)
if is_node_classifier:
train_loader, val_loader, test_loader = dataset, dataset, dataset
train_mask, val_mask, test_mask = get_test_val_train_mask(
dataset, split=split
)
else:
train_loader, val_loader, test_loader = generate_dataloaders(
dataset, args.dataset, args.batch_size, split=split
)
train_mask, val_mask, test_mask = None, None, None
training_config = TrainConfig(
lr=lr,
batch_size=1 if is_node_classifier else train_loader.batch_size,
epochs=args.epochs,
weight_decay=weight_decay
)
if args.save_models:
save_path=f'models/{args.name}_lr{lr}_wd{weight_decay}_do{dropout_rate}_s{split}.pt'
else:
save_path=None
model = model_factory(dropout_rate=dropout_rate)
run_stats = train_eval(
model,
training_config,
train_loader,
val_loader,
test_loader,
loss_function=cross_entropy,
metric_function=accuracy_function,
logger = epoch_logger,
train_mask=train_mask,
val_mask=val_mask,
test_mask=test_mask,
save_path=save_path,
)
for key, value in run_stats.items():
mean_stats[key].append(value)
print(lr, weight_decay, dropout_rate, split, run_stats)
run_stats = {**run_stats, **log_const}
run_logger.log(run_stats)
for key in mean_stats.keys():
mean_stats[key] = np.mean(mean_stats[key])
print(lr, weight_decay, dropout_rate, "avg", mean_stats)
log_const["split"] = "avg"
mean_stats = {**mean_stats, **log_const}
run_logger.log(mean_stats)
if __name__ == "__main__":
main()