-
Notifications
You must be signed in to change notification settings - Fork 2
/
experiment.py
122 lines (100 loc) · 4.73 KB
/
experiment.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
import os
from torch import optim
from models import BaseVAE
from typing import TypeVar
import pytorch_lightning as pl
import torchvision.utils as vutils
Tensor = TypeVar("torch.tensor")
class VAEXperiment(pl.LightningModule):
def __init__(self,
vae_model: BaseVAE,
params: dict) -> None:
super(VAEXperiment, self).__init__()
self.model = vae_model
self.params = params
self.curr_device = None
self.hold_graph = False
self.save_hyperparameters(self.params)
try:
self.hold_graph = self.params['retain_first_backpass']
except:
pass
def forward(self, x: Tensor, **kwargs) -> Tensor:
return self.model(x, **kwargs)
def training_step(self, batch, batch_idx, optimizer_idx=0):
real_img, labels = batch
self.curr_device = real_img.device
results = self.forward(real_img, labels=labels)
train_loss = self.model.loss_function(*results,
M_N=self.params['kld_weight'], # al_img.shape[0]/ self.num_train_imgs,
optimizer_idx=optimizer_idx,
batch_idx=batch_idx)
self.log_dict({key: val.item() for key, val in train_loss.items()}, sync_dist=True)
return train_loss['loss']
def validation_step(self, batch, batch_idx, optimizer_idx=0):
real_img, labels = batch
self.curr_device = real_img.device
results = self.forward(real_img, labels=labels)
val_loss = self.model.loss_function(*results,
M_N=1.0, # real_img.shape[0]/ self.num_val_imgs,
optimizer_idx=optimizer_idx,
batch_idx=batch_idx)
self.log_dict({f"val_{key}": val.item() for key, val in val_loss.items()}, sync_dist=True)
def on_validation_end(self) -> None:
self.sample_images()
def sample_images(self):
# Get sample reconstruction image
test_input, test_label = next(iter(self.trainer.datamodule.test_dataloader()))
test_input = test_input.to(self.curr_device)
test_label = test_label.to(self.curr_device)
# test_input, test_label = batch
recons = self.model.generate(test_input, labels=test_label)
vutils.save_image(recons.data,
os.path.join(self.logger.log_dir,
"Reconstructions",
f"recons_{self.logger.name}_Epoch_{self.current_epoch}.png"),
normalize=True,
nrow=12)
try:
samples = self.model.sample(144,
self.curr_device,
labels=test_label)
vutils.save_image(samples.cpu().data,
os.path.join(self.logger.log_dir,
"Samples",
f"{self.logger.name}_Epoch_{self.current_epoch}.png"),
normalize=True,
nrow=12)
except Warning:
pass
def configure_optimizers(self):
optims = []
scheds = []
optimizer = optim.Adam(self.model.parameters(),
lr=self.params['LR'],
weight_decay=self.params['weight_decay'])
optims.append(optimizer)
# Check if more than 1 optimizer is required (Used for adversarial training)
try:
if self.params['LR_2'] is not None:
optimizer2 = optim.Adam(getattr(self.model, self.params['submodel']).parameters(),
lr=self.params['LR_2'])
optims.append(optimizer2)
except:
pass
try:
if self.params['scheduler_gamma'] is not None:
scheduler = optim.lr_scheduler.ExponentialLR(optims[0],
gamma=self.params['scheduler_gamma'])
scheds.append(scheduler)
# Check if another scheduler is required for the second optimizer
try:
if self.params['scheduler_gamma_2'] is not None:
scheduler2 = optim.lr_scheduler.ExponentialLR(optims[1],
gamma=self.params['scheduler_gamma_2'])
scheds.append(scheduler2)
except:
pass
return optims, scheds
except:
return optims