-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathtrainer.py
179 lines (146 loc) · 8.54 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
import torch
import numpy as np
from utils.stats_manager import StatsManager
from utils.data_logs import save_logs_train, save_logs_eval
import os
class Trainer:
def __init__(self, network, train_dataloader, eval_dataloader, criterion, optimizer, lr_scheduler, config):
self.config = config
self.network = network
self.stats_manager = StatsManager(config)
self.train_dataloader = train_dataloader
self.eval_dataloader = eval_dataloader
self.criterion = criterion
self.optimizer = optimizer
self.lr_scheduler = lr_scheduler
self.best_metric = 0.0
def train_epoch(self, epoch):
running_loss = []
self.network.train()
for idx, (inputs, labels_depth, labels_distance, labels_magnitude) in enumerate(self.train_dataloader, 0):
inputs = inputs.to(self.config['device']).float()
labels_depth = labels_depth.to(self.config['device']).float()
labels_distance = labels_distance.to(self.config['device']).float()
labels_magnitude = labels_magnitude.to(self.config['device']).float()
self.optimizer.zero_grad()
pred_depth, pred_distance, pred_magnitude = self.network(inputs)
loss = self.criterion(pred_depth, pred_distance, pred_magnitude,
labels_depth, labels_distance, labels_magnitude)
loss.backward()
self.optimizer.step()
running_loss.append(loss.item())
if idx % self.config['print_loss'] == 0:
running_loss = np.mean(np.array(running_loss))
print(f'Training loss on iteration {idx} = {running_loss}')
save_logs_train(os.path.join(self.config['exp_path'], self.config['exp_name']),
f'Training loss on iteration {idx} = {running_loss}')
running_loss = []
def eval_net(self, epoch):
stats_pred_depth = []
stats_pred_distance = []
stats_pred_magnitude = []
stats_lbl_depth = []
stats_lbl_distance = []
stats_lbl_magnitude = []
running_eval_loss = 0.0
self.network.eval()
for idx, (inputs, labels_depth, labels_distance, labels_magnitude) in enumerate(self.eval_dataloader, 0):
inputs = inputs.to(self.config['device']).float()
labels_depth = labels_depth.to(self.config['device']).float()
labels_distance = labels_distance.to(self.config['device']).float()
labels_magnitude = labels_magnitude.to(self.config['device']).float()
with torch.no_grad():
pred_depth, pred_distance, pred_magnitude = self.network(inputs)
eval_loss = self.criterion(pred_depth, pred_distance, pred_magnitude,
labels_depth, labels_distance, labels_magnitude)
running_eval_loss += eval_loss.item()
stats_pred_depth.append(pred_depth.detach().cpu().numpy())
stats_pred_distance.append(pred_distance.detach().cpu().numpy())
stats_pred_magnitude.append(pred_magnitude.detach().cpu().numpy())
stats_lbl_depth.append(labels_depth.detach().cpu().numpy())
stats_lbl_distance.append(labels_distance.detach().cpu().numpy())
stats_lbl_magnitude.append(labels_magnitude.detach().cpu().numpy())
mean_depth_err, mean_distance_err, mean_magnitude_err = \
self.stats_manager.get_stats(pred_depth=stats_pred_depth, pred_distance=stats_pred_distance, pred_magnitude=stats_pred_magnitude,
lbl_depth=stats_lbl_depth, lbl_distance=stats_lbl_distance, lbl_magnitude=stats_lbl_magnitude)
running_eval_loss = running_eval_loss / len(self.eval_dataloader)
print(f'### Evaluation loss on epoch {epoch} = {running_eval_loss}, mean DEPTH error = {mean_depth_err}, '
f'mean DISTANCE error = {mean_distance_err}, mean MAGNITUDE error = {mean_magnitude_err}')
save_logs_eval(os.path.join(self.config['exp_path'], self.config['exp_name']),
f'### Evaluation loss on epoch {epoch} = {running_eval_loss}, mean DEPTH error = {mean_depth_err}, '
f'mean DISTANCE error = {mean_distance_err}, mean MAGNITUDE error = {mean_magnitude_err}')
if self.best_metric < mean_magnitude_err:
self.best_metric = mean_magnitude_err
self.save_net_state(None, best=True)
def train(self):
if self.config['resume_training'] is True:
checkpoint = torch.load(os.path.join(self.config['exp_path'], self.config['exp_name'], 'latest_checkpoint.pkl'),
map_location=self.config['device'])
self.network.load_state_dict(checkpoint['model_weights'])
self.optimizer.load_state_dict(checkpoint['optimizer'])
for i in range(1, self.config['train_epochs'] + 1):
print('Training on epoch ' + str(i))
self.train_epoch(i)
self.save_net_state(i, latest=True)
if i % self.config['eval_net_epoch'] == 0:
self.eval_net(i)
if i % self.config['save_net_epochs'] == 0:
self.save_net_state(i)
self.lr_scheduler.step()
def save_net_state(self, epoch, latest=False, best=False):
if latest is True:
path_to_save = os.path.join(self.config['exp_path'], self.config['exp_name'], f'latest_checkpoint.pkl')
to_save = {
'epoch': epoch,
'model_weights': self.network.state_dict(),
'optimizer': self.optimizer.state_dict()
}
torch.save(to_save, path_to_save)
elif best is True:
path_to_save = os.path.join(self.config['exp_path'], self.config['exp_name'], f'best_model.pkl')
to_save = {
'epoch': epoch,
'stats': self.best_metric,
'model_weights': self.network.state_dict()
}
torch.save(to_save, path_to_save)
else:
path_to_save = os.path.join(self.config['exp_path'], self.config['exp_name'], f'model_epoch_{epoch}.pkl')
torch.save(self.network, path_to_save)
def test_net(self, test_dataloader):
stats_pred_depth = []
stats_pred_distance = []
stats_pred_magnitude = []
stats_lbl_depth = []
stats_lbl_distance = []
stats_lbl_magnitude = []
running_loss = 0.0
self.network.eval()
for idx, (inputs, labels_depth, labels_distance, labels_magnitude) in enumerate(test_dataloader, 0):
inputs = inputs.to(self.config['device']).float()
labels_depth = labels_depth.to(self.config['device']).float()
labels_distance = labels_distance.to(self.config['device']).float()
labels_magnitude = labels_magnitude.to(self.config['device']).float()
with torch.no_grad():
pred_depth, pred_distance, pred_magnitude = self.network(inputs)
eval_loss = self.criterion(pred_depth, pred_distance, pred_magnitude,
labels_depth, labels_distance, labels_magnitude)
running_loss += eval_loss.item()
stats_pred_depth.append(pred_depth.detach().cpu().numpy())
stats_pred_distance.append(pred_distance.detach().cpu().numpy())
stats_pred_magnitude.append(pred_magnitude.detach().cpu().numpy())
stats_lbl_depth.append(labels_depth.detach().cpu().numpy())
stats_lbl_distance.append(labels_distance.detach().cpu().numpy())
stats_lbl_magnitude.append(labels_magnitude.detach().cpu().numpy())
mean_depth_err, mean_distance_err, mean_magnitude_err = \
self.stats_manager.get_stats(pred_depth=stats_pred_depth, pred_distance=stats_pred_distance,
pred_magnitude=stats_pred_magnitude,
lbl_depth=stats_lbl_depth, lbl_distance=stats_lbl_distance,
lbl_magnitude=stats_lbl_magnitude)
running_eval_loss = running_loss / len(test_dataloader)
stats_description = f'### Test loss = {running_eval_loss}, mean DEPTH error = {mean_depth_err}, \
mean DISTANCE error = {mean_distance_err}, mean MAGNITUDE error = {mean_magnitude_err}'
print(stats_description)
history = open(os.path.join(self.config['exp_path'], self.config['exp_name'], '__testStats__.txt'), "a")
history.write(stats_description)
history.close()