-
Notifications
You must be signed in to change notification settings - Fork 34
/
default_learner.py
183 lines (159 loc) · 7.06 KB
/
default_learner.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
from collections import OrderedDict
import numpy as np
import torch
import torch.nn.functional as F
from tqdm import tqdm
from confidnet.learners.learner import AbstractLeaner
from confidnet.utils import misc
from confidnet.utils.logger import get_logger
from confidnet.utils.metrics import Metrics
LOGGER = get_logger(__name__, level="DEBUG")
class DefaultLearner(AbstractLeaner):
def train(self, epoch):
self.model.train()
metrics = Metrics(
self.metrics, self.prod_train_len, self.num_classes
)
loss, len_steps, len_data = 0, 0, 0
# Training loop
loop = tqdm(self.train_loader)
for batch_id, (data, target) in enumerate(loop):
data, target = data.to(self.device), target.to(self.device)
self.optimizer.zero_grad()
output = self.model(data)
if self.task == "classification":
current_loss = self.criterion(output, target)
elif self.task == "segmentation":
current_loss = self.criterion(output, target.squeeze(dim=1))
current_loss.backward()
loss += current_loss
self.optimizer.step()
if self.task == "classification":
len_steps += len(data)
len_data = len_steps
elif self.task == "segmentation":
len_steps += len(data) * np.prod(data.shape[-2:])
len_data += len(data)
# Update metrics
confidence, pred = F.softmax(output, dim=1).max(dim=1, keepdim=True)
metrics.update(pred, target, confidence)
# Update the average loss
loop.set_description(f"Epoch {epoch}/{self.nb_epochs}")
loop.set_postfix(
OrderedDict(
{
"loss_nll": f"{(loss / len_data):05.4e}",
"acc": f"{(metrics.accuracy / len_steps):05.2%}",
}
)
)
loop.update()
# Eval on epoch end
scores = metrics.get_scores(split="train")
logs_dict = OrderedDict(
{
"epoch": {"value": epoch, "string": f"{epoch:03}"},
"lr": {
"value": self.optimizer.param_groups[0]["lr"],
"string": f"{self.optimizer.param_groups[0]['lr']:05.1e}",
},
"train/loss_nll": {
"value": loss / len_data,
"string": f"{(loss / len_data):05.4e}",
},
}
)
for s in scores:
logs_dict[s] = scores[s]
# Val scores
if self.val_loader is not None:
val_losses, scores_val = self.evaluate(self.val_loader, self.prod_val_len, split="val")
logs_dict["val/loss_nll"] = {
"value": val_losses["loss_nll"].item() / self.nsamples_val,
"string": f"{(val_losses['loss_nll'].item() / self.nsamples_val):05.4e}",
}
for sv in scores_val:
logs_dict[sv] = scores_val[sv]
# Test scores
test_losses, scores_test = self.evaluate(self.test_loader, self.prod_test_len, split="test")
logs_dict["test/loss_nll"] = {
"value": test_losses["loss_nll"].item() / self.nsamples_test,
"string": f"{(test_losses['loss_nll'].item() / self.nsamples_test):05.4e}",
}
for st in scores_test:
logs_dict[st] = scores_test[st]
# Print metrics
misc.print_dict(logs_dict)
# Save the model checkpoint
self.save_checkpoint(epoch)
# CSV logging
misc.csv_writter(path=self.output_folder / "logs.csv", dic=OrderedDict(logs_dict))
# Tensorboard logging
self.save_tb(logs_dict)
# Scheduler step
if self.scheduler:
self.scheduler.step()
def evaluate(
self, dloader, len_dataset, split="test", mode="normal", samples=50, verbose=False
):
self.model.eval()
metrics = Metrics(self.metrics, len_dataset, self.num_classes)
loss = 0
# Special case of mc-dropout
if mode == "mc_dropout":
self.model.keep_dropout_in_test()
LOGGER.info(f"Sampling {samples} times")
# Evaluation loop
loop = tqdm(dloader, disable=not verbose)
for batch_id, (data, target) in enumerate(loop):
data, target = data.to(self.device), target.to(self.device)
with torch.no_grad():
if mode == "mcp":
output = self.model(data)
if self.task == "classification":
loss += self.criterion(output, target)
elif self.task == "segmentation":
loss += self.criterion(output, target.squeeze(dim=1))
confidence, pred = F.softmax(output, dim=1).max(dim=1, keepdim=True)
elif mode == "tcp":
output = self.model(data)
if self.task == "classification":
loss += self.criterion(output, target)
elif self.task == "segmentation":
loss += self.criterion(output, target.squeeze(dim=1))
probs = F.softmax(output, dim=1)
pred = probs.max(dim=1, keepdim=True)[1]
labels_hot = misc.one_hot_embedding(
target, self.num_classes
).to(self.device)
# Segmentation special case
if self.task == "segmentation":
labels_hot = labels_hot.squeeze(1).permute(0, 3, 1, 2)
confidence, _ = (labels_hot * probs).max(dim=1, keepdim=True)
elif mode == "mc_dropout":
if self.task == "classification":
outputs = torch.zeros(
samples, data.shape[0], self.num_classes
).to(self.device)
elif self.task == "segmentation":
outputs = torch.zeros(
samples,
data.shape[0],
self.num_classes,
data.shape[2],
data.shape[3],
).to(self.device)
for i in range(samples):
outputs[i] = self.model(data)
output = outputs.mean(0)
if self.task == "classification":
loss += self.criterion(output, target)
elif self.task == "segmentation":
loss += self.criterion(output, target.squeeze(dim=1))
probs = F.softmax(output, dim=1)
confidence = (probs * torch.log(probs + 1e-9)).sum(dim=1) # entropy
pred = probs.max(dim=1, keepdim=True)[1]
metrics.update(pred, target, confidence)
scores = metrics.get_scores(split=split)
losses = {"loss_nll": loss}
return losses, scores