-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
188 lines (147 loc) · 6.62 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
181
182
183
184
185
186
187
188
import torch
from torch import nn
from einops import rearrange
from opt import get_opts
from models import GCN, SAGE
# datasets
from dataset import GraphDataset
from torch.utils.data import DataLoader
import torch.nn.functional as F
# optimizer
from torch.optim import Adam
from torch.optim.lr_scheduler import CosineAnnealingLR
from pytorch_lightning import LightningModule, Trainer
from pytorch_lightning.callbacks import TQDMProgressBar
from pytorch_lightning.loggers import TensorBoardLogger
from sampler import NeighborSampler
def get_learning_rate(optimizer):
for param_group in optimizer.param_groups:
return param_group['lr']
class GCNSystem(LightningModule):
def __init__(self, hparams):
super().__init__()
self.save_hyperparameters(hparams) # params -> self.parmas
# self.gcn = GCN(
# input_dim=hparams.input_dim,
# hidden_dim_ls=hparams.hidden_dim_ls,
# output_dim=hparams.output_dim)
def setup(self, stage=None):
self.dataset = GraphDataset(
hparams.graph_name, split="all", ratio=hparams.ratio)
self.train_dataset = GraphDataset(
hparams.graph_name, split="train", ratio=hparams.ratio)
self.val_dataset = GraphDataset(
hparams.graph_name, split="val", ratio=hparams.ratio)
self.edge_index = self.train_dataset.edge_index
self.hidden_layer_num = hparams.hidden_layer_num
self.sample_neighbor_num = hparams.sample_neighbor_num
self.train_idx = self.train_dataset.node_ids
self.val_idx = self.val_dataset.node_ids
self.sage_neighsampler_parameters = {
'num_layers':hparams.hidden_layer_num,
'hidden_channels':hparams.hidden_dim,
'dropout':0.0,
'batchnorm': False,
}
self.sage = SAGE(in_channels=self.dataset.d, out_channels=self.dataset.c, **self.sage_neighsampler_parameters)
def forward(self, x, adjs): # TODO
return self.sage(x, adjs)
def train_dataloader(self):
return NeighborSampler(self.edge_index,
node_idx=self.train_idx,
sizes=[self.sample_neighbor_num]*self.hidden_layer_num,
batch_size=self.hparams.batch_size,
shuffle=True,
num_workers=4
)
# return DataLoader(self.train_dataset,
# shuffle=True,
# num_workers=4,
# batch_size=self.hparams.batch_size,
# pin_memory=True)
def val_dataloader(self):
return NeighborSampler(self.edge_index,
node_idx=self.val_idx,
sizes=[self.sample_neighbor_num]*self.hidden_layer_num,
batch_size=self.hparams.batch_size,
shuffle=False,
num_workers=4
)
# return DataLoader(self.val_dataset,
# shuffle=False,
# num_workers=4,
# batch_size=self.hparams.batch_size,
# pin_memory=True)
def configure_optimizers(self):
self.opt = Adam(self.sage.parameters(), lr=self.hparams.lr)
scheduler = CosineAnnealingLR(self.opt, hparams.num_epochs, hparams.lr/1e2)
return [self.opt], [scheduler]
def training_step(self, batch, batch_idx): # TODO:batch_idx 第几个batch
# batch 就是 getitem出来的东西，是个dict
# 只负责写loss
batch_size, n_id, adjs = batch
y_true = self.dataset.y[n_id[:batch_size]]
# self调用forward函数
output = self(self.dataset.x[n_id], adjs)
output = F.log_softmax(output, dim=1)
loss = F.nll_loss(output, y_true)
pred = output.argmax(dim=1)
acc = sum(pred == y_true)/pred.shape[0]
self.log('lr', self.opt.param_groups[0]['lr'])
self.log('train/loss', loss)
self.log('train/acc', acc, prog_bar=True)
return loss
def validation_step(self, batch, batch_idx):
batch_size, n_id, adjs = batch
y_true = self.dataset.y[n_id[:batch_size]]
# self调用forward函数
output = self(self.dataset.x[n_id], adjs)
output = F.log_softmax(output, dim=1)
loss = F.nll_loss(output, y_true)
pred = output.argmax(dim=1)
acc = sum(pred == y_true)/pred.shape[0]
log = {'val_loss': loss,
'val_acc': acc}
return log
def validation_epoch_end(self, outputs):
mean_loss = torch.tensor([x['val_loss'] for x in outputs]).mean()
mean_acc = torch.tensor([x['val_acc'] for x in outputs]).mean()
self.log('val/loss', mean_loss, prog_bar=True)
self.log('val/acc', mean_acc, prog_bar=True)
if __name__ == '__main__':
hparams = get_opts()
system = GCNSystem(hparams)
pbar = TQDMProgressBar(refresh_rate=1)
callbacks = [pbar]
logger = TensorBoardLogger(save_dir="logs",
name=hparams.exp_name,
default_hp_metric=False)
trainer = Trainer(max_epochs=hparams.num_epochs,
callbacks=callbacks,
logger=logger,
enable_model_summary=True,
accelerator='auto',
devices=1,
num_sanity_val_steps=0,
log_every_n_steps=1,
check_val_every_n_epoch=20,
benchmark=True)
trainer.fit(system)
# test
# ckpt_path = "./logs/exp/version_0/checkpoints/epoch=19-step=3140.ckpt"
# # system = CoordMLPSystem(hparams)
# system = CoordMLPSystem.load_from_checkpoint(ckpt_path)
# # trainer = Trainer()
# # 自动恢复模型,可以继续训练
# # trainer.fit(model, ckpt_path=ckpt_path)
# mlp = system.mlp
# mlp.eval()
# image = imageio.imread("/Users/sutongtong/Documents/GitHub/Coordinate-MLPs/images/fox.jpg")
# resolution = 128
# pred = torch.zeros([resolution, resolution, 3])
# for i in range(resolution):
# for j in range(resolution):
# rgb = mlp(torch.FloatTensor([[i/resolution, j/resolution]]))
# pred[i, j, :] = rgb
# print(pred.shape)
# imageio.imwrite("./images/output.jpg", pred.detach().numpy())