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
|
require 'torch'
require 'nn'
if opt.type == 'cuda' then
require 'cunn'
end
----------------------------------------------------------------------
print(sys.COLORS.red .. '==> construct CNN')
-- Raw training set
-- This gets around 70.6 % on test starting around epoc 115
-- 73.6 % on training
--
-- With horizontal flipping
-- This gets around 74.5 % on test starting around epoc 156
-- 72.1 % on training
--
-- th run.lua -r 1.48242 --sgdLearningRateDecay 0.00195 --sgdMomentum 0.09982 --sgdWeightDecay 2.46e-05 -s full
local CNN = nn.Sequential()
CNN:add(nn.SpatialConvolution(3, 10, 3, 3, 1, 1, 1, 1))
CNN:add(nn.ReLU())
CNN:add(nn.SpatialDropout(0.2))
CNN:add(nn.SpatialMaxPooling(2,2,2,2))
CNN:add(nn.SpatialBatchNormalization(10))
CNN:add(nn.SpatialConvolution(10, 20, 3, 3, 1, 1, 1, 1))
CNN:add(nn.ReLU())
CNN:add(nn.SpatialDropout(0.2))
CNN:add(nn.SpatialMaxPooling(2,2,2,2))
CNN:add(nn.SpatialBatchNormalization(20))
CNN:add(nn.SpatialConvolution(20, 40, 3, 3, 1, 1, 1, 1))
CNN:add(nn.ReLU())
CNN:add(nn.SpatialDropout(0.25))
CNN:add(nn.SpatialBatchNormalization(40))
CNN:add(nn.View(40*8*8))
local classifier = nn.Sequential()
classifier:add(nn.Linear(40*8*8, 500))
classifier:add(nn.BatchNormalization(500))
classifier:add(nn.ReLU())
classifier:add(nn.Dropout(0.5))
classifier:add(nn.Linear(500, 10))
classifier:add(nn.LogSoftMax())
local model = nn.Sequential()
model:add(CNN)
model:add(classifier)
-- Loss: NLL
local loss = nn.ClassNLLCriterion()
----------------------------------------------------------------------
print(sys.COLORS.red .. '==> here is the CNN:')
print(model)
if opt.type == 'cuda' then
model:cuda()
loss:cuda()
end
-- return package:
return {
model = model,
loss = loss,
}
|
