【Julia Deep Learning CV】 第一篇 MNIST

                                                                                                                                                               工欲善其事必先有其器

                                                                                                                                                    我们先还是老老实实做调包侠吧

1、安装必要的包：

        Pkg.add("Statistics")

        Pkg.add("Flux")

2、先上代码：

using Statistics
using Flux: onehotbatch, onecold, crossentropy, throttle
using Base.Iterators: repeated
imgs = Flux.Data.MNIST.images()
labels = Flux.Data.MNIST.labels();

 #以上代码就是调用包，加载imgs和labels

imgs[27454]

 

labels[27454] 
# 7

X = hcat(float.(reshape.(imgs, :))...) #MNIST图片是28 * 28 （长， 宽）二维，变成成一维
Y = onehotbatch(labels, 0:9) #对Y进行编码

model = Chain( Dense(28^2, 32, relu), Dense(32, 10), softmax) #神经网络一共三层：第一层，28^2  ->  32；第二层，32 -                                                                                                               #>10； 第三层，softmax

#以下是优化过程

loss(x, y) = crossentropy(m(x), y) 
opt = ADAM(); 

accuracy(x, y) = mean(onecold(m(x)) .== onecold(y)) 

dataset = repeated((X,Y),200) 
evalcb = () -> @show(loss(X, Y)) 

#可以开始训练了

Flux.train!(loss, params(m), dataset, opt, cb = throttle(evalcb, 10));

#训练结果如下：

loss(X, Y) = 2.3259583f0 (tracked)
loss(X, Y) = 1.6830894f0 (tracked)
loss(X, Y) = 1.1227762f0 (tracked)
loss(X, Y) = 0.7927527f0 (tracked)
loss(X, Y) = 0.6152953f0 (tracked)
loss(X, Y) = 0.51356655f0 (tracked)
loss(X, Y) = 0.44959342f0 (tracked)
loss(X, Y) = 0.4059622f0 (tracked)
loss(X, Y) = 0.3741082f0 (tracked)
loss(X, Y) = 0.3512681f0 (tracked)
loss(X, Y) = 0.33128205f0 (tracked)
loss(X, Y) = 0.31474704f0 (tracked)
loss(X, Y) = 0.3016968f0 (tracked)
loss(X, Y) = 0.28936785f0 (tracked)
loss(X, Y) = 0.27849576f0 (tracked)
loss(X, Y) = 0.2688136f0 (tracked)
......

accuracy(X, Y) #训练集上的结果
#0.92735

test_X = hcat(float.(reshape.(Flux.Data.MNIST.images(:test), :))...)
test_Y = onehotbatch(Flux.Data.MNIST.labels(:test), 0:9);
model(test_X[:,5287])
accuracy(test_X, test_Y) #测试集上的结果
#0.924

代码比较简单，所用到的mlp也很简单，loss用的是交叉熵，梯度用的是ADAM。

第二篇我们还是用这些扩展一下，总不能一老做调包侠吧