/
train.py
333 lines (287 loc) · 9.67 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
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
from preprocessing import data_loader
from model import CNN
import torch
import torch.nn as nn
import torch.optim as optim
import numpy as np
import matplotlib.pyplot as plt
from tqdm import tqdm
from preprocessing import AlexEmbed
from torch.utils.data import DataLoader
alexembed = AlexEmbed()
def loaders(dataset, split = 0.8, batch_size = 256):
train_size = int(split * len(dataset))
val_size = len(dataset) - train_size
train_set, val_set = torch.utils.data.random_split(dataset, [train_size, val_size])
train_loader = DataLoader(train_set, batch_size, shuffle=True)
val_loader = DataLoader(val_set, 1024, shuffle=True)
return train_loader, val_loader
def train(net, train_loader, val_loader, batch_size=64, lr=0.001, num_epochs=30):
# Fixed PyTorch random seed for reproducible result
torch.manual_seed(0)
if torch.cuda.is_available():
net = net.cuda()
# cross entropy loss function and adaptive moment estimation optimizer
criterion = nn.CrossEntropyLoss()
optimizer = optim.AdamW(net.parameters(), lr = lr, weight_decay=0.1)
# softmax for predictions
softmax = nn.Softmax(dim = 1)
# initialize error and loss history
train_err = np.zeros(num_epochs)
train_loss = np.zeros(num_epochs)
val_err = np.zeros(num_epochs)
val_loss = np.zeros(num_epochs)
for epoch in tqdm(range(num_epochs)):
total_train_loss = 0.0
total_train_err = 0.0
train_iters = 0
total_val_loss = 0.0
total_val_err = 0.0
val_iters = 0
train_batches = 0
net.train()
for batch in train_loader:
train_batches += 1
imgs, labels = batch.values()
if torch.cuda.is_available():
imgs = imgs.cuda()
labels = labels.cuda()
optimizer.zero_grad()
outputs = net(imgs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
pred = softmax(outputs)
#print("debug1", pred.shape, labels.shape)
# find error and loss for training data
#total_train_err += (np.argmax(pred.detach().cpu(), 1) != np.argmax(labels.cpu(), 1)).sum().item()
total_train_err += (np.argmax(pred.detach().cpu(), 1) != labels.cpu()).sum().item()
total_train_loss += loss.item()
train_iters += len(labels)
val_batches = 0
net.eval()
for batch in val_loader:
val_batches += 1
imgs, labels = batch.values()
if torch.cuda.is_available():
imgs = imgs.cuda()
labels = labels.cuda()
outputs = net(imgs)
loss = criterion(outputs, labels)
pred = softmax(outputs)
# find error and loss for training data
total_val_err += (np.argmax(pred.detach().cpu(), 1) != labels.cpu()).sum().item()
total_val_loss += loss.item()
val_iters += len(labels)
# record the average error (per iteration) and loss (per batch) for each epoch
train_err[epoch] = total_train_err / train_iters
train_loss[epoch] = total_train_loss / train_batches
val_err[epoch] = total_val_err / val_iters
val_loss[epoch] = total_val_loss / val_batches
print(f"Epoch {epoch}: Train err: {train_err[epoch]} Val err: {val_err[epoch]} Train loss: {train_loss[epoch]} Val loss: {val_loss[epoch]}")
# save model
model_path = "bs{}_lr{}_epoch{}".format(batch_size,
lr,
epoch)
torch.save(net.state_dict(), model_path)
return train_err, train_loss, val_err, val_loss
def evaluation(net, batch_size=1024): # Evaluate the error/loss of a loaded model
# Fixed PyTorch random seed for reproducible result
torch.manual_seed(0)
if torch.cuda.is_available():
#print("cuda activated")
net = net.to('cuda')
train_loader, val_loader = data_loader(batch_size=batch_size)
# cross entropy loss function
criterion = nn.CrossEntropyLoss()
# softmax for predictions
softmax = nn.Softmax(dim = 1)
# initialize error and loss history
total_train_loss = 0.0
total_train_err = 0.0
train_iters = 0
total_val_loss = 0.0
total_val_err = 0.0
val_iters = 0
train_batches = 0
net.eval()
for batch in tqdm(train_loader):
train_batches += 1
imgs, labels = batch.values()
if torch.cuda.is_available():
imgs = imgs.to('cuda')
labels = labels.to('cuda')
outputs = net(imgs)
loss = criterion(outputs, labels)
pred = softmax(outputs)
# find error and loss for training data
total_train_err += (np.argmax(pred.detach().cpu(), 1) != np.argmax(labels.cpu(), 1)).sum().item()
print(total_train_err)
total_train_loss += loss.item()
train_iters += len(labels)
val_batches = 0
for batch in tqdm(val_loader):
val_batches += 1
imgs, labels = batch.values()
if torch.cuda.is_available():
imgs = imgs.to('cuda')
labels = labels.to('cuda')
outputs = net(imgs)
loss = criterion(outputs, labels)
pred = softmax(outputs)
# find error and loss for training data
total_val_err += (np.argmax(pred.detach().cpu(), 1) != np.argmax(labels.cpu(), 1)).sum().item()
total_val_loss += loss.item()
val_iters += len(labels)
return total_train_loss, total_train_err/len(train_loader), total_val_loss, total_val_err/len(val_loader)
def plot(train_err, train_loss, val_err, val_loss):
n = len(train_err) # number of epochs
fig, (ax1, ax2) = plt.subplots(1, 2)
ax1.set_title("Train vs Validation Error")
ax1.plot(range(1,n+1), train_err, label="Train")
ax1.plot(range(1,n+1), val_err, label="Validation")
ax1.set_xlabel("Epoch")
ax1.set_ylabel("Error")
ax1.legend(loc='best')
ax1.xaxis.get_major_locator().set_params(integer=True)
ax2.set_title("Train vs Validation Loss")
ax2.plot(range(1,n+1), train_loss, label="Train")
ax2.plot(range(1,n+1), val_loss, label="Validation")
ax2.set_xlabel("Epoch")
ax2.set_ylabel("Loss")
ax2.legend(loc='best')
ax2.xaxis.get_major_locator().set_params(integer=True)
plt.show()
def performance_per_class(net):
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
net = net.to(device)
net.eval()
_, val_loader = data_loader(batch_size=1)
errors = {
0:0,
1:0,
2:0,
3:0,
4:0,
5:0,
6:0,
7:0,
8:0,
9:0,
10:0,
11:0,
12:0,
13:0,
14:0,
}
total = {
0:0,
1:0,
2:0,
3:0,
4:0,
5:0,
6:0,
7:0,
8:0,
9:0,
10:0,
11:0,
12:0,
13:0,
14:0,
}
wrong_guesses = {
0:0,
1:0,
2:0,
3:0,
4:0,
5:0,
6:0,
7:0,
8:0,
9:0,
10:0,
11:0,
12:0,
13:0,
14:0,
}
guesses = {
0:0,
1:0,
2:0,
3:0,
4:0,
5:0,
6:0,
7:0,
8:0,
9:0,
10:0,
11:0,
12:0,
13:0,
14:0,
}
softmax = nn.Softmax(dim = 1)
for batch in val_loader:
img, label = batch.values()
if torch.cuda.is_available():
img = img.to('cuda')
label = label.to('cuda')
output = softmax(net(img))
pred = np.argmax(output.detach().cpu()).item()
#print(pred)
truth = np.argmax(label.cpu()).item()
if pred != truth:
errors[truth] += 1
wrong_guesses[pred] += 1
total[truth] += 1
guesses[pred] += 1
for i in range(15):
if guesses[i] == 0:
wrong_guesses[i] = guesses[i]
else:
wrong_guesses[i] /= guesses[i]
if total[i] == 0:
errors[i] = 0
else:
errors[i] /= total[i]
return errors, wrong_guesses, guesses
if __name__ == "__main__":
# net = Baseline()
# train_err, train_loss, val_err, val_loss = train(net, 64, 0.001, 20)
# plot(train_err, train_loss, val_err, val_loss)
#net = CNN()
#net.load_state_dict(torch.load("./models/bs256_lr0.0001_epoch29", map_location=torch.device('cpu')))
#print(net)
#train_err, train_loss, val_err, val_loss = train(net, 128, 0.0001, 29)
#error_rate, wrong_guess_rate, guesses = performance_per_class(net)
# print("The error rate is:"+ str(error_rate))
# print(wrong_guess_rate)
# print(guesses)
x = np.array([0,1,2,3,4,5,6,7,8,9,10,11,12,13,14])
xtick = ["sitting", "using_laptop", "hugging", "sleeping", "drinking", "clapping", "dancing", "cycling",
"calling", "laughing", "eating", "fighting", "listening_to_music", "running", "texting"]
plt.xticks(x, xtick, rotation=45)
plt.plot(x, error_rate.values())
plt.title("Error rates per class")
plt.xlabel("Class")
plt.ylabel("Error rate")
plt.show()
plt.xticks(x, xtick, rotation=45)
plt.plot(x, wrong_guess_rate.values())
plt.title("Wrong guess rate per class")
plt.xlabel("Class")
plt.ylabel("Wrong guess rate")
plt.show()
plt.xticks(x, xtick, rotation=45)
plt.plot(x, guesses.values())
plt.title("Guesses per class")
plt.xlabel("Class")
plt.ylabel("Number of guesses")
plt.show()
#from torchinfo import summary
#summary(net, input_size=(256, 3, 224, 224))