-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrainer.py
211 lines (151 loc) · 7.78 KB
/
trainer.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
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
import torch
from torch import nn
import numpy as np
import os
from matplotlib import pyplot as plt
from timer import Timer
class Trainer():
def __init__(self, args_dict):
self.args_dict = args_dict
self.timer = Timer()
def calc_metrics(self, pred, target):
_, pred = torch.max(pred, dim = 1)
correct = (pred == target).sum()
return correct/target.size(0)
def save_model(self, model_name, dataset_name, epoch, model):
folder_name = 'weights-' + str(model_name) + '-' + str(dataset_name)
save_dir = os.path.join(self.args_dict['weights_save_path'], folder_name)
if os.path.isdir(save_dir) == False:
os.mkdir(save_dir)
save_name = 'Epoch-' + str(epoch) + '.pth'
torch.save(model.state_dict(), os.path.join(save_dir, save_name))
def load_model(self, model_name, dataset_name, epoch, model):
folder_name = 'weights-' + str(model_name) + '-' + str(dataset_name)
save_dir = os.path.join(self.args_dict['weights_save_path'], folder_name)
load_name = 'Epoch-' + str(epoch) + '.pth'
model.load_state_dict(torch.load(os.path.join(save_dir, load_name)))
return model
def calculate_no_of_parameters(self, model):
return sum(p.numel() for p in model.parameters() if p.requires_grad)
def eval_single(self):
raise NotImplementedError('Eval single is not implemented')
def train(self, model, train_loader, optimizer, epoch):
loss_array = []
accuracy_array = []
epoch_time = []
timer = Timer()
device = self.args_dict['device']
file_path = self.args_dict['record_save_path']
model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target.to(device)
timer.start()
optimizer.zero_grad()
output = model(data)
loss = nn.CrossEntropyLoss()(output, target)
loss.backward()
optimizer.step()
timer.stop()
loss_array.append(loss.item())
accuracy = self.calc_metrics(output, target)
accuracy_array.append(accuracy.item())
epoch_time.append(timer.get_elapsed_time())
if batch_idx % 10 == 0:
print(f'Train Epoch: {epoch} [{batch_idx * len(data)}/{len(train_loader.dataset)} ({100. * batch_idx / len(train_loader):.0f}%)]\tLoss: {loss.item():.6f}')
print(f'Train Epoch: {epoch} Loss: {np.mean(loss_array)} Accuracy: {np.mean(accuracy_array)}')
with open(file_path, 'a') as f:
f.write(f'Epoch: {epoch}')
f.write('\n')
f.write(f'Train Loss: {np.mean(loss_array)}')
f.write('\n')
f.write(f'Train Accuracy: {np.mean(accuracy_array)}')
f.write('\n')
f.write(f'Epoch Time: {np.sum(epoch_time)}')
f.write('\n')
f.write("Average time per iteration: " + str(np.mean(epoch_time)))
f.write('\n')
f.write("-------------------------------------------------\n")
return np.mean(loss_array), np.mean(accuracy_array)
def train_models(self):
for m in range(self.args_dict['num_models']):
dataset_index = m % self.args_dict['num_datasets']
dataset_name = self.args_dict[('dataset_name', dataset_index)]
model = self.args_dict[('model', m)]
model_name = self.args_dict[('model_name', m)] #model_inf[1]
optimizer = self.args_dict[('optimizers', m)]
scheduler = self.args_dict[('schedulers', m)]
with open(self.args_dict['record_save_path'], 'a') as f:
f.write(f'Model name: ----------- {model_name} --------------')
f.write('\n')
f.write(f'Dataset name: ----------- {dataset_name} --------------\n')
print(f'Model name: ----------- {model_name} --------------')
train_losses = []
train_accuracies = []
test_losses = []
test_accuracies = []
for epoch in range(self.args_dict['epochs']):
train_loss, train_accuracy = self.train(model=model, train_loader=self.args_dict[('trainloader', dataset_index)], optimizer=optimizer, epoch=epoch)
test_loss, test_accuracy = self.evaluate(model, self.args_dict[('testloader', dataset_index)], scheduler)
self.save_model(model_name, dataset_name, epoch, model)
train_accuracies.append(train_accuracy)
train_losses.append(train_loss)
test_accuracies.append(test_accuracy)
test_losses.append(test_loss)
self.plot_results(train_losses, train_accuracies, test_losses, test_accuracies, model_name, dataset_name)
def plot_results(self, train_losses, train_accuracies, test_losses, test_accuracies, model_name, dataset_name):
plt.figure()
plt.plot(train_losses, label='Train Loss')
plt.plot(test_losses, label='Test Loss')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.title('Loss vs Epoch')
plt.legend()
save_dir = 'graphs'
if os.path.isdir(save_dir) == False:
os.mkdir(save_dir)
path = os.path.join(save_dir, f'Loss-{model_name}-{dataset_name}.png')
plt.savefig(path)
plt.figure()
plt.plot(train_accuracies, label='Train Accuracy')
plt.plot(test_accuracies, label='Test Accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.title('Accuracy vs Epoch')
plt.legend()
path = os.path.join(save_dir, f'Accuracy-{model_name}-{dataset_name}.png')
plt.savefig(path)
def evaluate(self, model, test_loader, scheduler):
loss_array = []
accuracy_array = []
eval_time = []
model.eval()
for data in test_loader:
inputs, labels = data
inputs = inputs.to(self.args_dict['device'])
labels = labels.to(self.args_dict['device'])
with torch.no_grad():
self.timer.start()
outputs = model(inputs)
self.timer.stop()
eval_time.append(self.timer.get_elapsed_time())
loss = self.args_dict['loss_function'](outputs, labels)
metrics = self.calc_metrics(outputs, labels)
metrics = metrics.cpu().numpy()
loss_array.append(loss.item())
accuracy_array.append(metrics.item())
if scheduler is not None:
scheduler.step()
print(f'Validation Loss: {np.mean(loss_array)}')
print(f'Validation Accuracy: {np.mean(accuracy_array)}')
print(f'Validation Time: {np.sum(eval_time)}')
with open(self.args_dict['record_save_path'], 'a') as f:
f.write(f'Validation Loss: {np.mean(loss_array)}')
f.write('\n')
f.write(f'Validation Accuracy: {np.mean(accuracy_array)}')
f.write('\n')
f.write(f'Validation Time: {np.sum(eval_time)}')
f.write('\n')
f.write("Average time per iteration: " + str(np.mean(eval_time)))
f.write('\n')
f.write("-------------------------------------------------\n")
return np.mean(loss_array), np.mean(accuracy_array)