3、Spark MLlib Deep Learning Convolution Neural Network(深度学习-卷积神经网络)3.3

http://blog.csdn.net/sunbow0

第三章Convolution Neural Network (卷积神经网络)

3实例

3.1 测试数据

按照上例数据，或者新建图片识别数据。

3.2 CNN实例

//2 测试数据

Logger.getRootLogger.setLevel(Level.WARN)

valdata_path="/user/tmp/deeplearn/train_d.txt"

valexamples=sc.textFile(data_path).cache()

valtrain_d1=examples.map{ line =>

valf1 = line.split("\t")

valf =f1.map(f =>
f.toDouble)

valy =f.slice(0,10)

valx =f.slice(10,f.length)

(newBDM(1,y.length,
y), (new BDM(1,x.length,
x)).reshape(28,28) /
255.0)

}

valtrain_d=train_d1.map(f=> (f._1, f._2))

//3 设置训练参数，建立模型

// opts:迭代步长，迭代次数，交叉验证比例

valopts= Array(100.0,1.0,0.0)

train_d.cache

valnumExamples=train_d.count()

println(s"numExamples = $numExamples.")

valCNNmodel=newCNN().

setMapsize(new BDM(1,2, Array(28.0,28.0))).

setTypes(Array("i",
"c","s","c","s")).

setLayer(5).

setOnum(10).

setOutputmaps(Array(0.0,
6.0,0.0,12.0,0.0)).

setKernelsize(Array(0.0,
5.0,0.0,5.0,0.0)).

setScale(Array(0.0,
0.0,2.0,0.0,2.0)).

setAlpha(1.0).

setBatchsize(50.0).

setNumepochs(1.0).

CNNtrain(train_d,opts)

//4 模型测试

valCNNforecast=CNNmodel.predict(train_d)

valCNNerror=CNNmodel.Loss(CNNforecast)

println(s"NNerror = $CNNerror.")

valprintf1=CNNforecast.map(f=> (f.label.data(0),
f.predict_label.data(0))).take(200)

println("预测结果——实际值：预测值：误差")

for(i <-0 until
printf1.length)

println(printf1(i)._1 +"\t"
+printf1(i)._2 +"\t" + (printf1(i)._2
-printf1(i)._1))   val
numExamples = train_d.count()

println(s"numExamples = $numExamples.")

println(mynn._2)

for(i <-0 to
mynn._1.length -1) {

print(mynn._1(i) +"\t")

}

println()

println("mynn_W1")

valtmpw1=mynn._3(0)

for(i <-0 to
tmpw1.rows -1) {

for(j <-0 to
tmpw1.cols -1) {

print(tmpw1(i,j) +
"\t")

}

println()

}

valNNmodel=newNeuralNet().

setSize(mynn._1).

setLayer(mynn._2).

setActivation_function("sigm").

setOutput_function("sigm").

setInitW(mynn._3).

NNtrain(train_d,nnopts)

//5 NN模型测试

valNNforecast=NNmodel.predict(train_d)

valNNerror=NNmodel.Loss(NNforecast)

println(s"NNerror = $NNerror.")

valprintf1=NNforecast.map(f=> (f.label.data(0),
f.predict_label.data(0))).take(200)

println("预测结果——实际值：预测值：误差")

for(i <-0 until
printf1.length)

println(printf1(i)._1 +"\t"
+printf1(i)._2 +"\t" + (printf1(i)._2
-printf1(i)._1))

转载请注明出处：

http://blog.csdn.net/sunbow0

版权声明：本文为博主原创文章，未经博主允许不得转载。