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

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