-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
207 lines (178 loc) · 7.61 KB
/
train.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
import torch
import sys
import copy
import os
from data_utils import *
from model import *
from utils import *
def train(model, tr_dataset, val_dataset, criterion, optimizer, sav_path='./checkpoints/temp.pth', num_epochs=25, bs=32, device='cuda:0'):
model = model.to(device)
best_loss = 100000.0
best_dice = 0
print("Training parameters: \n----------")
print("batch size: ", bs)
print("num epochs: ", num_epochs)
for epoch in range(num_epochs):
print(f'Epoch {epoch}/{num_epochs - 1}')
print('-' * 10)
bs_count=0
inputs_li, labels_li, text_ids_li, text_li = [], [], [], []
running_loss = 0
running_dice = 0
count = 0
#run training
# print("eere: ",len(tr_dataset))
for i in range(len(tr_dataset)):
inputs, labels,_, text = tr_dataset[i]
inputs_li.append(inputs)
labels_li.append(labels)
text_li = text_li + [text]*(inputs.shape[0])
bs_count += 1
if (bs_count%bs==0) or (i==len(tr_dataset)-1):
#start training
bs_count=0
inputs = torch.cat(inputs_li,dim=0)
labels = torch.cat(labels_li, dim=0)
inputs = inputs.to(device)
labels = labels.to(device)
with torch.set_grad_enabled(True):
optimizer.zero_grad()
outputs = model(inputs, text_li)
seg_loss=0
for c in criterion:
seg_loss += c(outputs, labels.float())
seg_loss.backward()
optimizer.step()
running_loss += seg_loss.cpu()
preds = (outputs>=0.5)
ri, ru = running_stats(labels,preds)
running_dice += dice_collated(ri,ru)
count += ri.shape[0]
inputs_li = []
labels_li = []
text_li = []
epoch_dice = running_dice / count
print("Training loss: ", running_loss/(1+(len(tr_dataset)//bs)))
print("Training dice: ", epoch_dice)
#do val if epoch is a multiple of 5
if epoch%5==0:
running_dice = 0
count=0
for i in range(len(val_dataset)):
inputs, labels,_, text = val_dataset[i]
inputs_li.append(inputs)
labels_li.append(labels)
text_li = text_li + [text]*(inputs.shape[0])
bs_count += 1
if bs_count%bs==0:
#start training
bs_count=0
inputs = torch.cat(inputs_li,dim=0)
labels = torch.cat(labels_li, dim=0)
inputs = inputs.to(device)
labels = labels.to(device)
with torch.set_grad_enabled(False):
outputs = model(inputs, text_li)
preds = (outputs>=0.5)
ri, ru = running_stats(labels,preds)
running_dice += dice_collated(ri,ru)
count += ri.shape[0]
inputs_li = []
labels_li = []
text_li = []
# epoch_dice = running_dice / (len(val_dataset))
epoch_dice = running_dice / count
print(f'Val Dice: {epoch_dice:.4f}')
# deep copy the model
if epoch_dice > best_dice:
# best_loss = epoch_loss
best_dice = epoch_dice
torch.save(model.state_dict(),sav_path)
return model
def train_dl(model, dataloaders, dataset_sizes, criterion, optimizer, scheduler, sav_path='./checkpoints/temp.pth', num_epochs=25, bs=32, device='cuda:0'):
model = model.to(device)
best_dice = 0
best_loss=10000
print("Training parameters: \n----------")
print("batch size: ", bs)
print("num epochs: ", num_epochs)
for epoch in range(num_epochs):
print(f'Epoch {epoch}/{num_epochs - 1}')
print('-' * 10)
# Each epoch has a training and validation phase
for phase in ['train', 'val']:
if phase == 'train':
model.train() # Set model to training mode
else:
model.eval() # Set model to evaluate mode
running_loss = 0.0
running_intersection = 0
running_union = 0
running_corrects = 0
running_dice = 0
intermediate_count = 0
count = 0
preds_all = []
gold = []
# Iterate over data.
for inputs, labels,text_idxs, text in dataloaders[phase]:
count+=1
intermediate_count += inputs.shape[0]
inputs = inputs.to(device)
labels = labels.to(device)
# zero the parameter gradients
optimizer.zero_grad()
# forward
# track history if only in train
with torch.set_grad_enabled(phase == 'train'):
outputs = model(inputs, text)
# print(outputs)
# print(outputs.shape)
# print(outputs)
loss=0
seg_loss = 0
for c in criterion:
try:
seg_loss += c(outputs, text, labels.float())
except:
seg_loss += c(outputs, labels.float())
loss += seg_loss
# backward + optimize only if in training phase
if phase == 'train':
loss.backward()
optimizer.step()
with torch.no_grad():
preds = (outputs>=0.5)
# preds_all.append(preds.cpu())
# gold.append(labels.cpu())
# epoch_dice = dice_coef(preds,labels)
# if count%100==0:
# print('iteration dice: ', epoch_dice)
# statistics
running_loss += loss.item() * inputs.size(0)
ri, ru = running_stats(labels,preds)
running_dice += dice_collated(ri,ru)
# if count%5==0:
# print(count)
# print(running_loss, intermediate_count)
# print(running_loss/intermediate_count)
if phase == 'train':
scheduler.step()
print("all 0 sanity check for preds: ", preds.any())
print("all 1 sanity check for preds: ", not preds.all())
epoch_loss = running_loss / ((dataset_sizes[phase]))
epoch_dice = running_dice / ((dataset_sizes[phase]))
# epoch_dice = dice_coef(torch.cat(preds_all,axis=0),torch.cat(gold,axis=0))
print(f'{phase} Loss: {epoch_loss:.4f} Dice: {epoch_dice:.4f}')
# deep copy the model
if phase == 'val' and epoch_loss < best_loss:
best_loss = epoch_loss
best_dice = epoch_dice
torch.save(model.state_dict(),sav_path)
elif phase == 'val' and np.isnan(epoch_loss):
print("nan loss but saving model")
torch.save(model.state_dict(),sav_path)
print(f'Best val loss: {best_loss:4f}, best val accuracy: {best_dice:2f}')
# load best model weights
# model.load_state_dict(best_model_wts)
return model