-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
164 lines (138 loc) · 6.11 KB
/
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
158
159
160
161
162
163
164
import os
import fire
import numpy as np
import torch
from torch import optim
from torch.utils.tensorboard import SummaryWriter
from torchvision import datasets, transforms
from libs.Visualize import Visualize
from models.VAE import VAE
class Main():
def __init__(self, z_dim):
"""Constructor
Args:
z_dim (int): Dimensions of the latent variable.
Returns:
None.
"""
self.z_dim = z_dim
self.dataloader_train = None
self.dataloader_valid = None
self.dataloader_test = None
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.model = VAE(self.z_dim).to(self.device)
self.writer = SummaryWriter(log_dir="./logs")
self.lr = 0.001
self.optimizer = optim.Adam(self.model.parameters(), lr=self.lr)
self.num_max_epochs = 1000
self.num_no_improved = 0
self.num_batch_train = 0
self.num_batch_valid = 0
self.loss_valid = 10 ** 7 # Initialize with a large value
self.loss_valid_min = 10 ** 7 # Initialize with a large value
self.Visualize = Visualize(self.z_dim, self.dataloader_test, self.model, self.device)
def createDirectories(self):
"""Create directories for the tensorboard and learned model
Args:
None.
Returns:
None.
"""
if not os.path.exists("./logs"):
os.makedirs("./logs")
if not os.path.exists("./params"):
os.makedirs("./params")
def createDataLoader(self):
"""Download MNIST and convert it to data loaders
Args:
None.
Returns:
None.
"""
transform = transforms.Compose([transforms.ToTensor(), transforms.Lambda(lambda x: x.view(-1))]) # Preprocessing for MNIST images
dataset_train_valid = datasets.MNIST("./", train=True, download=True, transform=transform) # Separate train data and test data to get a dataset
dataset_test = datasets.MNIST("./", train=False, download=True, transform=transform)
# Use 20% of train data as validation data
size_train_valid = len(dataset_train_valid) # 60000
size_train = int(size_train_valid * 0.8) # 48000
size_valid = size_train_valid - size_train # 12000
dataset_train, dataset_valid = torch.utils.data.random_split(dataset_train_valid, [size_train, size_valid])
# Create dataloaders from the datasets
self.dataloader_train = torch.utils.data.DataLoader(dataset_train, batch_size=1000, shuffle=True)
self.dataloader_valid = torch.utils.data.DataLoader(dataset_valid, batch_size=1000, shuffle=False)
self.dataloader_test = torch.utils.data.DataLoader(dataset_test, batch_size=1000, shuffle=False)
self.Visualize.dataloader_test = self.dataloader_test
def train_batch(self):
"""Batch-based learning for training data
Args:
None.
Returns:
None.
"""
self.model.train()
for x, _ in self.dataloader_train:
lower_bound, _, _ = self.model(x, self.device)
loss = -sum(lower_bound)
self.model.zero_grad()
loss.backward()
self.optimizer.step()
self.writer.add_scalar("Loss_train/KL", -lower_bound[0].cpu().detach().numpy(), self.num_iter + self.num_batch_train)
self.writer.add_scalar("Loss_train/Reconst", -lower_bound[1].cpu().detach().numpy(), self.num_iter + self.num_batch_train)
self.num_batch_train += 1
self.num_batch_train -= 1
def valid_batch(self):
"""Batch-based learning for validating data
Args:
None.
Returns:
None.
"""
loss = []
self.model.eval()
for x, _ in self.dataloader_valid:
lower_bound, _, _ = self.model(x, self.device)
loss.append(-sum(lower_bound).cpu().detach().numpy())
self.writer.add_scalar("Loss_valid/KL", -lower_bound[0].cpu().detach().numpy(), self.num_iter + self.num_batch_valid)
self.writer.add_scalar("Loss_valid/Reconst", -lower_bound[1].cpu().detach().numpy(), self.num_iter + self.num_batch_valid)
self.num_batch_valid += 1
self.num_batch_valid -= 1
self.loss_valid = np.mean(loss)
self.loss_valid_min = np.minimum(self.loss_valid_min, self.loss_valid)
def early_stopping(self):
"""Judging early stopping
Args:
None.
Returns:
None.
"""
if self.loss_valid_min < self.loss_valid: # If the loss of this iteration is greater than the minimum loss of the previous iterations, the counter variable is incremented.
self.num_no_improved += 1
print(f"Validation got worse for the {self.num_no_improved} time in a row.")
else: # If the loss of this iteration is the same or smaller than the minimum loss of the previous iterations, reset the counter variable and save parameters.
self.num_no_improved = 0
torch.save(self.model.state_dict(), f"./params/model_z_{self.z_dim}.pth")
def main(self):
self.createDirectories()
self.createDataLoader()
print("-----Start training-----")
for self.num_iter in range(self.num_max_epochs):
self.train_batch()
self.valid_batch()
print(f"[EPOCH{self.num_iter + 1}] loss_valid: {int(self.loss_valid)} | Loss_valid_min: {int(self.loss_valid_min)}")
self.early_stopping()
if self.num_no_improved >= 10:
print("Apply early stopping")
break
self.writer.close()
print("-----Stop training-----")
print("-----Start Visualization-----")
self.model.load_state_dict(torch.load(f"./params/model_z_{self.z_dim}.pth"))
self.model.eval()
self.Visualize.createDirectories()
self.Visualize.reconstruction()
self.Visualize.latent_space()
self.Visualize.lattice_point()
self.Visualize.walkthrough()
print("-----Stop Visualization-----")
if __name__ == '__main__':
fire.Fire(Main)