-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_hybrid.py
175 lines (142 loc) · 7.44 KB
/
train_hybrid.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
import argparse
import json
import os
from pprint import pprint
import torch
import torch.nn as nn
import torch.optim as optim
from sklearn.metrics import accuracy_score
import numpy as np
import random
import matplotlib.pyplot as plt
from os.path import join
from utils import load_data, stack_windowed_data, check_max
from dataset_loaders import create_dataloaders
from models import HybridModel
seed = 7777777
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
if torch.cuda.is_available():
device = 'cuda'
torch.cuda.manual_seed_all(seed)
else:
device = 'cpu'
def train(model: torch.nn.Module, train_loader, val_loader, criterion, optimizer, num_epochs,
model_name: str = 'bilstm-model') -> dict[str, list]:
best_val_loss = float('inf')
best_epoch = 0
history = {"train_loss": [], "train_acc": [], "val_loss": [], "val_acc": []}
for epoch in range(num_epochs):
train_losses, train_acc = [], []
model.train()
for padded_sequences, lengths, labels in train_loader:
padded_sequences = padded_sequences.to(device)
optimizer.zero_grad()
outputs = model(padded_sequences.float().to(device), lengths)
loss = criterion(outputs.view(-1, 4), labels.long().view(-1).to(device))
loss.backward()
optimizer.step()
train_losses.append(loss.item())
_, predicted = torch.max(outputs.view(-1, 4), 1)
accuracy = accuracy_score(labels.view(-1).numpy(), predicted.cpu().detach().numpy())
train_acc.append(accuracy)
avg_train_loss = np.mean(train_losses)
history['train_loss'].append(avg_train_loss)
avg_train_accuracy = np.mean(train_acc)
history['train_acc'].append(avg_train_accuracy)
model.eval()
val_losses, val_acc = [], []
with torch.no_grad():
for padded_sequences, lengths, labels in val_loader:
outputs = model(padded_sequences.float().to(device), lengths)
loss = criterion(outputs.view(-1, 4), labels.long().view(-1).to(device))
val_losses.append(loss.item())
_, predicted = torch.max(outputs.view(-1, 4), 1)
accuracy = accuracy_score(labels.view(-1).numpy(), predicted.cpu().detach().numpy())
val_acc.append(accuracy)
avg_val_loss = np.mean(val_losses)
history['val_loss'].append(avg_val_loss)
avg_val_accuracy = np.mean(val_acc)
history['val_acc'].append(avg_val_accuracy)
if epoch % 10 == 0:
print(f'Epoch [{epoch}/{num_epochs}]: Train Loss: {round(avg_train_loss, 3)}, '
f'Validation Loss: {round(avg_val_loss, 3)}, Val Accuracy: {round(avg_val_accuracy, 3)}')
if avg_val_loss < best_val_loss:
best_epoch = epoch
best_val_loss = avg_val_loss
torch.save(model, os.path.join('models', f'{model_name}.pt'))
with open(os.path.join('models', 'history', f'{model_name}.json'), 'w', encoding='utf-8') as f:
json.dump(history, f, ensure_ascii=False, indent=4)
fig, axs = plt.subplots(ncols=2, figsize=(10, 4))
axs[0].plot(history['train_loss'], label='Train Loss')
axs[0].plot(history['val_loss'], label='Validation Loss')
axs[0].set_xlabel('Epoch #')
axs[0].set_ylabel('Loss')
axs[1].plot(history['train_acc'], label='Train Accuracy')
axs[1].plot(history['val_acc'], label='Validation Accuracy')
axs[1].set_xlabel('Epoch #')
axs[1].set_ylabel('Accuracy')
plt.tight_layout()
plt.legend()
plt.savefig(os.path.join('models', 'plots', f'{model_name}.png'))
plt.close()
print(f'* Epoch [{epoch}/{num_epochs}]: Train Loss: {round(avg_train_loss, 3)}, '
f'Validation Loss: {round(avg_val_loss, 3)}, Val Accuracy: {round(avg_val_accuracy, 3)}')
print('Best epoch:', best_epoch, 'with validation loss:', best_val_loss)
return history
def main(args):
data = load_data(file_path=join('data', 'W_AUGMENTED_DATA.json'))
sequences, targets = stack_windowed_data(data)
train_loader, val_loader, test_loader, class_weights = create_dataloaders(sequences, targets,
batch_size=args.batch_size,
return_class_weights=True,
verbose=False)
model = HybridModel(
n_conv_blocks=args.n_conv_blocks,
cnn_out_channels=args.cnn_out_channels,
cnn_dropout=args.cnn_dropout,
lstm_num_layers=args.lstm_num_layers,
lstm_hidden_size=args.lstm_hidden_size,
lstm_bidirectional=args.lstm_bidirectional,
lstm_dropout=args.lstm_dropout
).to(device)
class_weights_tensor = torch.FloatTensor(class_weights).to(device)
criterion = nn.CrossEntropyLoss(weight=class_weights_tensor)
optimizer = optim.Adam(model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay)
history = train(model, train_loader, val_loader, criterion, optimizer, num_epochs=args.num_epochs,
model_name=args.model_name)
print('Model trained. Final validation accuracy:', history['val_acc'][-1])
if __name__ == '__main__':
parser = argparse.ArgumentParser(description="Train a model with given parameters.")
parser.add_argument("--n_conv_blocks", type=check_max, default=2,
help="Number of convolutional blocks in the model. Maximum of 7.")
parser.add_argument("--cnn_out_channels", type=int, default=32,
help="Output size of the convolution layers in the model.")
parser.add_argument("--cnn_dropout", type=float, default=0.388,
help="Dropout probability between the CNN blocks.")
parser.add_argument("--lstm_bidirectional", action="store_true",
help="If set, the model will be bidirectional.")
parser.add_argument("--lstm_num_layers", type=int, default=3,
help="Number of layers in the LSTM.")
parser.add_argument("--lstm_hidden_size", type=int, default=174,
help="Size of a hidden layer in the LSTM.")
parser.add_argument("--lstm_dropout", type=float, default=0.2689664397393872,
help="Dropout probability between the LSTM layers.")
parser.add_argument("--learning_rate", type=float, default=0.009728535043244946,
help="Learning rate for training the model.")
parser.add_argument("--weight_decay", type=float, default=7e-05,
help="Weight decay in the optimizer.")
parser.add_argument("--batch_size", type=float, default=32,
help="Batch size for training the model.")
parser.add_argument("--num_epochs", type=int, default=250,
help="Number of epochs to train the model.")
parser.add_argument("--verbose", action="store_true",
help="If set, will print the information to the terminal.")
parser.add_argument("--model_name", type=str, default=None,
help="Name of the model where the weights will be saved.")
args = parser.parse_args()
print("Training model with parameters:")
pprint(vars(args))
print("-" * 30)
main(args)