-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
164 lines (146 loc) · 6.66 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
import os
import torch
import pandas as pd
import numpy as np
from torch.utils.data import DataLoader
from dgllife.utils import EarlyStopping
from preprocessing import load_dataset, dataset_split
from model import Model
from utils import collate, get_loss, get_metrics, set_seed, \
plot_training_curve, checkpoint
def train_one_epoch(args, epoch, device, model,
data_loader, score, optimizer, criterion):
model.train()
labels, preds, losses = [], [], []
for batch_id, batch_data in enumerate(data_loader):
smiles, graphs, label = batch_data
graphs = graphs.to(device)
label = torch.tensor(label).float().to(device)
node_feat = graphs.ndata.pop('node_feat')
edge_feat = graphs.edata.pop('edge_feat')
coord_feat = graphs.ndata.pop('coord_feat')
# logits: predicted results, ins_pred: predicted results for conformers, attn: attn coef for conformers
logits, ins_pred, attn = model(graphs, node_feat, edge_feat, coord_feat)
logits = logits.squeeze()
loss = criterion(logits.squeeze(), ins_pred.squeeze(dim=-1), label)
if args.task == 'classification':
logits = torch.sigmoid(logits)
losses.append(loss.item())
optimizer.zero_grad()
loss.backward()
optimizer.step()
labels.extend(label.detach().cpu().numpy().tolist())
if logits.shape == 0:
preds.extend([logits.item()])
else:
try:
preds.extend(logits.detach().cpu().numpy().tolist())
except:
print(logits)
preds.extend([logits.item()])
score['loss'] += [np.mean(losses)]
score['epoch'] += [epoch]
metrics = get_metrics(labels, preds, args.metric)
score['metric_1'] += [metrics[0]]
score['metric_2'] += [metrics[1]]
def eval_one_epoch(args, device, model, data_loader, score):
model.eval()
smiles_list, label_list, pred_list = [], [], []
for batch_id, batch_data in enumerate(data_loader):
smiles, graphs, label = batch_data
graphs = graphs.to(device)
label = torch.tensor(label).float().to(device)
node_feat = graphs.ndata.pop('node_feat')
edge_feat = graphs.edata.pop('edge_feat')
coord_feat = graphs.ndata.pop('coord_feat')
logits, ins_pred, attn = model(graphs, node_feat, edge_feat, coord_feat)
logits = logits.squeeze()
if args.task == 'classification':
logits = torch.sigmoid(logits)
smiles_list.extend(smiles)
label_list.extend(label.detach().cpu().numpy().tolist())
if logits.shape == 0:
pred_list.extend([logits.item()])
else:
try:
pred_list.extend(logits.detach().cpu().numpy().tolist())
except:
print(logits)
pred_list.extend([logits.item()])
metrics = get_metrics(label_list, pred_list, args.metric)
score['val_metric_1'] += [metrics[0]]
score['val_metric_2'] += [metrics[1]]
return [smiles_list, pred_list, label_list]
def Train(args, logger):
if not os.path.exists(args.save_path):
os.makedirs(args.save_path)
if args.device_id != 'cpu':
logger.info('Training on GPU')
device = torch.device('cuda:{}'.format(args.device_id))
else:
logger.info('Training on CPU')
device = torch.device('cpu')
argsDict = args.__dict__
with open(os.path.join(args.save_path, 'setting.txt'), 'w') as f:
for eachArg, value in argsDict.items():
f.writelines(eachArg + ' : ' + str(value) + '\n')
logger.info("Configuration: {}".format(args))
set_seed(args.seed)
dataset = load_dataset(args)
train, val, test = dataset_split(dataset, args)
logger.info(f"Train dataset: {len(train)}; Val dataset: {len(val)}; Test dataset: {len(test)}")
train_loader = DataLoader(train, batch_size=args.batch_size, collate_fn=collate,
shuffle=True, drop_last=False)
if len(val) >= 2:
val_loader = DataLoader(val, batch_size=args.batch_size, collate_fn=collate,
shuffle=False, drop_last=False)
else:
val_loader = None
test_loader = DataLoader(test, batch_size=args.batch_size, collate_fn=collate,
shuffle=False, drop_last=False)
model = Model(in_size=dataset[0][1].ndata['node_feat'].shape[-1],
hidden_size=args.hidden_size,
edge_feat_size=dataset[0][1].edata['edge_feat'].shape[-1],
num_layer=args.num_layer,
dropout=args.dropout).to(device)
optimizer = torch.optim.Adam(model.parameters(),
lr=args.learning_rate,
weight_decay=args.weight_decay)
criterion = get_loss(args)
stopper = EarlyStopping(patience=30,
filename=os.path.join(args.save_path, 'model.pkl'),
metric='roc_auc_score')
score = {'loss': [], 'epoch': [],
'metric_1': [], 'metric_2': [],
'val_metric_1': [], 'val_metric_2': []}
for epoch in range(args.epoch):
train_one_epoch(args, epoch, device, model,
train_loader, score, optimizer, criterion)
if epoch % args.print_every == 0:
if val_loader:
_ = eval_one_epoch(args, device, model, val_loader, score)
else:
_ = eval_one_epoch(args, device, model, test_loader, score)
logger.info("Epoch {}/{}: loss: {:.3f}, train {}: {:.3f}, {}: {:.3f} "
" val {}: {:.3f}, {}: {:.3f}".format(
epoch + 1, args.epoch, score['loss'][-1],
args.metric[0], score['metric_1'][-1],
args.metric[1], score['metric_2'][-1],
args.metric[0], score['val_metric_1'][-1],
args.metric[1], score['val_metric_2'][-1]))
m = checkpoint(args, model, score, args.metric)
if m:
best_model = m
early_stop = stopper.step(score['val_metric_1'][-1], model)
# if early_stop:
# break
# stopper.load_checkpoint(model)
plot_training_curve(args, score)
test_result = eval_one_epoch(args, device, best_model, test_loader, score)
logger.info("Test {}: {:.3f}, {}: {:.3f}".format(
args.metric[0], score['val_metric_1'][-1],
args.metric[1], score['val_metric_2'][-1]))
pd.DataFrame(np.array(test_result).T,
columns=['SMILES', 'Predict', 'Label']
).to_csv(os.path.join(args.save_path, 'predict.csv'), index=False)
logger.info(f'Training done, file saved in {args.save_path}')