-
Notifications
You must be signed in to change notification settings - Fork 0
/
neuro_simple_v1.jl
172 lines (142 loc) · 4.69 KB
/
neuro_simple_v1.jl
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
# original version with operations changed to support Julia 1.1,
# + some minor changes to variable naming, data manipulation, etc.
using LinearAlgebra
using MLDatasets
using Random
struct network_v1
num_layers::Int64
sizearr::Array{Int64,1}
biases::Array{Array{Float64,1},1}
weights::Array{Array{Float64,2},1}
zs::Array{Array{Float64,1},1}
activations::Array{Array{Float64,1},1}
∇_b::Array{Array{Float64,1},1}
∇_w::Array{Array{Float64,2},1}
δ_∇_b::Array{Array{Float64,1},1}
δ_∇_w::Array{Array{Float64,2},1}
δs::Array{Array{Float64,1},1}
end
σ(z) = 1/(1+exp(-z))
σ_grad(z) = σ(z)*(1-σ(z))
function (net::network_v1)(a)
for (w, b) in zip(net.weights, net.biases)
a = σ.(w*a + b)
end
return a
end
function network_v1(sizes)
num_layers = length(sizes)
sizearr = sizes
biases = [randn(y) for y in sizes[2:end]]
weights = [randn(y, x) for (x, y) in zip(sizes[1:end-1], sizes[2:end])]
zs = [zeros(y) for y in sizes[2:end]]
activations = [zeros(y) for y in sizes[1:end]]
∇_b = [zeros(y) for y in sizes[2:end]]
∇_w = [zeros(y, x) for (x, y) in zip(sizes[1:end-1], sizes[2:end])]
δ_∇_b = [zeros(y) for y in sizes[2:end]]
δ_∇_w = [zeros(y, x) for (x, y) in zip(sizes[1:end-1], sizes[2:end])]
δs = [zeros(y) for y in sizes[2:end]]
network_v1(num_layers, sizearr, biases, weights, zs, activations,∇_b,∇_w,δ_∇_b,δ_∇_w,δs)
end
function update_batch(net::network_v1, batch, η)
∇_b = net.∇_b
∇_w = net.∇_w
for i in 1:length(∇_b)
∇_b[i] .= 0.0
end
for i in 1:length(∇_w)
∇_w[i] .= 0.0
end
δ_∇_b = net.δ_∇_b
δ_∇_w = net.δ_∇_w
for (x, y) in batch
backprop!(net, x, y)
for i in 1:length(∇_b)
∇_b[i] .+= δ_∇_b[i]
end
for i in 1:length(∇_w)
∇_w[i] .+= δ_∇_w[i]
end
end
coef = (η/length(batch))
for i in 1:length(∇_b)
net.biases[i] .-= coef.*∇_b[i]
end
for i in 1:length(∇_w)
net.weights[i] .-= coef.*∇_w[i]
end
end
function backprop!(net::network_v1, x, y)
∇_b = net.δ_∇_b
∇_w = net.δ_∇_w
len = net.num_layers - 1
activations = net.activations
activations[1] .= x
zs = net.zs
δs = net.δs
for i in 1:len
b, w, z = net.biases[i], net.weights[i], zs[i]
mul!(z,w,activations[i]) # z = w * inp
z .+= b
activations[i+1] .= σ.(z)
end
δ = δs[end]
δ .= (activations[end] .- y) .* σ_grad.(zs[end])
∇_b[end] .= δ
mul!(∇_w[end], δ, activations[end-1]') # ∇_w[end] = δ * activations[end-1]'
for l in 1:len-1
z = zs[end-l]
mul!(δs[end-l], net.weights[end-l+1]', δ) # δs[end-l] = net.weights[end-l+1]' * δ
δ = δs[end-l]
δ .*= σ_grad.(z)
∇_b[end-l] .= δ
mul!(∇_w[end-l], δ, activations[end-l-1]') # ∇_w[end-l] = δ * activations[end-1-l]'
end
return nothing
end
function SGDtrain(net::network_v1, traindata, epochs, batch_size, η, testdata=nothing)
n_test = testdata != nothing ? length(testdata) : nothing
n = length(traindata)
traindata = shuffle(traindata) # one time shuffle for performance, then only take random batch index
println("========")
for j in 1:epochs
idx = randperm(n ÷ batch_size) .* batch_size
batches = [traindata[k-batch_size+1 : k] for k in idx]
@time for batch in batches
update_batch(net, batch, η)
end
if testdata != nothing
println("Epoch ", j,": ", evaluate(net, testdata), "/", n_test)
else
println("Epoch ", j," complete.")
end
end
end
function evaluate(net::network_v1, testdata)
test_results = [(findmax(net(x))[2] - 1, y) for (x, y) in testdata]
return sum(Int(x == y) for (x, y) in test_results)
end
function loaddata(rng = 1:60000)
train_x, train_y = MNIST.traindata(Float64, Vector(rng))
train_x = [train_x[:,:,x][:] for x in 1:size(train_x, 3)]
train_y = [vectorize(x) for x in train_y]
traindata = [(x, y) for (x, y) in zip(train_x, train_y)]
test_x, test_y = MNIST.testdata(Float64)
test_x = [test_x[:,:,x][:] for x in 1:size(test_x, 3)]
testdata = [(x, y) for (x, y) in zip(test_x, test_y)]
return traindata, testdata
end
function vectorize(n)
ev = zeros(10)
ev[n+1] = 1
return ev
end
function main_v1()
epochs = 10
batch_size = 10
η = 1.25
net = network_v1([784, 30, 10])
traindata, testdata = loaddata()
SGDtrain(net, traindata, epochs, batch_size, η, testdata)
# @profiler SGDtrain(net, traindata, 1, batch_size, η, testdata)
end