-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathgraph_main.py
158 lines (117 loc) · 5.02 KB
/
graph_main.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
import os
import yaml
import time
import math
import copy
import random
import argparse
import numpy as np
import pandas as pd
from functools import partial
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader
from graph_dataset_pyg import PygGraphPropPredDataset
from utils import init_params, seed_everything, collate_basis, collate_basis_sign
from model import MoleculeEncoder, EigenMLP_BN, SpaSpeGraph
from evaluation import ogbg_evaluation
from ogb.graphproppred import Evaluator
def train_epoch(model, train_loader, optimizer):
model.train()
batch_loss = 0.
batch_acc = 0.
for (spa_batch, spe_batch) in train_loader:
spa_batch = spa_batch.to(device)
spe_batch = spe_batch.to(device)
optimizer.zero_grad()
loss = model(spa_batch, spe_batch)
batch_loss += loss.item()
loss.backward()
optimizer.step()
return batch_loss / len(train_loader)
def eval_epoch(model, valid_loader):
model.eval()
emb = []
y_true = []
for (spa_batch, spe_batch) in valid_loader:
spa_batch = spa_batch.to(device)
spe_batch = spe_batch.to(device)
with torch.no_grad():
_, spa = model.spa_encoder(spa_batch.batch, spa_batch.x, spa_batch.edge_index, spa_batch.edge_attr)
_, spe = model.spe_encoder(spe_batch.batch, spe_batch.x, spe_batch.edge_index, spe_batch.edge_attr)
spa, spe = spa.detach().cpu(), spe.detach().cpu()
emb.append((spa + spe) / 2.)
y_true.append(spa_batch.y.detach().cpu())
emb = torch.cat(emb, dim=0).numpy()
y_true = torch.cat(y_true, dim=0).numpy()
return emb, y_true
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--cuda', type=int, default=0)
parser.add_argument('--dataset', type=str, default='molesol')
args = parser.parse_args()
config = yaml.load(open('config.yaml'), Loader=yaml.SafeLoader)[args.dataset]
print(config)
seed_everything(args.seed)
device = 'cuda:{}'.format(args.cuda)
dataset_name = 'ogbg-{}'.format(args.dataset)
epoch = config['epoch']
lr = config['lr']
weight_decay = config['weight_decay']
bs = config['batch_size']
period = config['period']
hidden_dim = config['hidden_dim']
pooling = config['pooling']
spa_layer = config['spa_layer']
spe_layer = config['spe_layer']
spa_dropout = config['spa_dropout']
spe_dropout = config['spe_dropout']
t = config['t']
dataset = PygGraphPropPredDataset(name = dataset_name)
split_idx = dataset.get_idx_split()
evaluator = Evaluator(dataset_name)
task_type = dataset.task_type
num_tasks = dataset.num_tasks
train_loader = DataLoader(dataset, batch_size=bs, shuffle=True, collate_fn=partial(collate_basis_sign, period=period))
valid_loader = DataLoader(dataset, batch_size=bs, shuffle=False, collate_fn=partial(collate_basis_sign, period=period))
#train_loader = DataLoader(dataset, batch_size=bs, shuffle=True, collate_fn=partial(collate_pyg, period=period))
#valid_loader = DataLoader(dataset, batch_size=bs, shuffle=False, collate_fn=partial(collate_pyg, period=period))
spa_encoder = MoleculeEncoder(hidden_dim, spa_layer, spa_dropout, pooling)
spe_encoder = EigenMLP_BN(period, hidden_dim, spe_layer, spe_dropout, pooling)
model = SpaSpeGraph(spa_encoder, spe_encoder, hidden_dim, t).to(device)
evaluator = Evaluator(dataset_name)
optimizer = torch.optim.Adam(model.parameters(), lr=lr, weight_decay=weight_decay)
valid_curve = []
test_curve = []
train_curve = []
print('training')
emb, y_true = eval_epoch(model, valid_loader)
train_score, val_score, test_score = ogbg_evaluation(emb, y_true, split_idx, evaluator, task_type, num_tasks)
print('Before training: ', train_score, val_score, test_score)
for i in range(epoch):
model.train()
batch_loss = 0.
batch_acc = 0.
for (spa_batch, spe_batch) in train_loader:
spa_batch = spa_batch.to(device)
spe_batch = spe_batch.to(device)
optimizer.zero_grad()
loss = model(spa_batch, spe_batch)
batch_loss += loss.item()
loss.backward()
optimizer.step()
emb, y_true = eval_epoch(model, valid_loader)
train_score, val_score, test_score = ogbg_evaluation(emb, y_true, split_idx, evaluator, task_type, num_tasks)
train_curve.append(train_score)
valid_curve.append(val_score)
test_curve.append(test_score)
print(batch_loss/len(train_loader), train_score, val_score, test_score)
if 'classification' in dataset.task_type:
best_val_epoch = np.argmax(np.array(valid_curve))
best_train = max(train_curve)
else:
best_val_epoch = np.argmin(np.array(valid_curve))
best_train = min(train_curve)
print('final: ', best_val_epoch, valid_curve[best_val_epoch], test_curve[best_val_epoch])