-
Notifications
You must be signed in to change notification settings - Fork 55
/
train.py
180 lines (127 loc) · 5.04 KB
/
train.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
from __future__ import annotations
# type: ignore
import argparse
from collections import namedtuple
from timeit import default_timer
from typing import Callable
import torch
import torch.nn as nn
import torch.nn.functional as F
from munch import Munch
from tqdm import tqdm
from utils import utils
from matgl.models import MEGNet
from matgl.models.helper import MLP
def train(
model: nn.Module,
device: torch.device,
optimizer: torch.optim.Optimizer,
loss_function: Callable[[torch.Tensor, torch.Tensor], torch.Tensor],
data: tuple,
dataloader: tuple,
):
model.train()
avg_loss = 0
start = default_timer()
for g, labels in tqdm(dataloader):
optimizer.zero_grad()
g = g.to(device)
labels = labels.to(device)
node_feat = torch.hstack((g.ndata["attr"], g.ndata["pos"]))
edge_feat = g.edata["edge_attr"]
attrs = torch.ones(g.batch_size, 2).to(device) * torch.tensor([data.z_mean, data.num_bond_mean]).to(device)
pred = model(g, edge_feat, node_feat, attrs)
loss = loss_function(pred, (labels - data.mean) / data.std)
loss.backward()
optimizer.step()
avg_loss += loss.detach()
stop = default_timer()
avg_loss = avg_loss.cpu().item() / len(dataloader)
epoch_time = stop - start
return avg_loss, epoch_time
def validate(
model: nn.Module,
device: torch.device,
loss_function: Callable[[torch.Tensor, torch.Tensor], torch.Tensor],
data: namedtuple,
dataloader: namedtuple,
):
avg_loss = 0
start = default_timer()
with torch.no_grad():
for g, labels in dataloader:
g = g.to(device)
labels = labels.to(device)
node_feat = torch.hstack((g.ndata["attr"], g.ndata["pos"]))
edge_feat = g.edata["edge_attr"]
attrs = torch.ones(g.batch_size, 2).to(device) * torch.tensor([data.z_mean, data.num_bond_mean]).to(device)
pred = model(g, edge_feat, node_feat, attrs)
loss = loss_function(data.mean + pred * data.std, labels)
avg_loss += loss
stop = default_timer()
avg_loss = avg_loss.cpu().item() / len(dataloader)
epoch_time = stop - start
return avg_loss, epoch_time
def run(
args: argparse.ArgumentParser,
config: Munch,
data: namedtuple,
):
g_sample = data.train[0][0]
node_feat = torch.hstack((g_sample.ndata["attr"], g_sample.ndata["pos"]))
edge_feat = g_sample.edata["edge_attr"]
attrs = torch.tensor([data.z_mean, data.num_bond_mean])
node_embed = MLP([node_feat.shape[-1], config.model.DIM])
edge_embed = MLP([edge_feat.shape[-1], config.model.DIM])
attr_embed = MLP([attrs.shape[-1], config.model.DIM])
device = torch.device(config.model.device if torch.cuda.is_available() else "cpu")
model = MEGNet(
in_dim=config.model.DIM,
num_blocks=config.model.num_blocks,
hiddens=[config.model.N1, config.model.N2],
conv_hiddens=[config.model.N1, config.model.N1, config.model.N2],
s2s_num_layers=config.model.s2s_num_layers,
s2s_num_iters=config.model.s2s_num_iters,
output_hiddens=[config.model.N2, config.model.N3],
is_classification=False,
node_embed=node_embed,
edge_embed=edge_embed,
attr_embed=attr_embed,
)
model = model.to(device)
print(model)
optimizer = torch.optim.Adam(model.parameters(), config.optimizer.lr)
train_loss_function = F.mse_loss
validate_loss_function = F.l1_loss
dataloaders = utils.create_dataloaders(config, data)
logger = utils.StreamingJSONWriter(filename="./qm9_logs.json")
print("## Training started ##")
for epoch in tqdm(range(config.optimizer.max_epochs)):
train_loss, train_time = train(model, device, optimizer, train_loss_function, data, dataloaders.train)
val_loss, val_time = validate(model, device, validate_loss_function, data, dataloaders.val)
print(
f"Epoch: {epoch + 1:03} Train Loss: {train_loss:.4f} "
f"Val Loss: {val_loss:.4f} Train Time: {train_time:.2f} s. "
f"Val Time: {val_time:.2f} s."
)
log_dict = {
"Epoch": epoch + 1,
"train_loss": train_loss,
"val_loss": val_loss,
"train_time": train_time,
"val_time": val_time,
}
logger.dump(log_dict)
print("## Training finished ##")
if __name__ == "__main__":
argparser = argparse.ArgumentParser("Agent Backbone Training")
argparser.add_argument("--config-name", default="qm9_test", type=str)
argparser.add_argument("--test-validation", dest="test_validation", action="store_true")
argparser.add_argument("--no-test-validation", dest="test_validation", action="store_false")
argparser.set_defaults(test_validation=True)
argparser.add_argument("--seed", default=0, type=int)
args = argparser.parse_args()
utils.set_seed(args.seed)
config = utils.prepare_config(f"./configs/{args.config_name}.yaml")
data = utils.prepare_data(config)
run(args, config, data)