Package org.apache.sysds.runtime.matrix.data

Class LibMatrixDNN

java.lang.Object
org.apache.sysds.runtime.matrix.data.LibMatrixDNN
public class LibMatrixDNN extends Object

  • Constructor Details

    • LibMatrixDNN

      public LibMatrixDNN()

  • Method Details

    • conv2d

      public static void conv2d(MatrixBlock input, MatrixBlock filter, MatrixBlock outputBlock, DnnParameters params)
      This method performs convolution (i.e. cross-correlation) operation on input
      Parameters:
      input - input batch
      filter - filter
      outputBlock - output of convolution
      params - convolution parameters

    • conv2dBackwardData

      public static void conv2dBackwardData(MatrixBlock filter, MatrixBlock dout, MatrixBlock outputBlock, DnnParameters params)
      This method computes the backpropogation errors for previous layer of convolution operation
      Parameters:
      filter - filter used in conv2d
      dout - errors from next layer
      outputBlock - output errors
      params - convolution parameters

    • conv2dBackwardFilter

      public static void conv2dBackwardFilter(MatrixBlock input, MatrixBlock dout, MatrixBlock outputBlock, DnnParameters params)
      This method computes the backpropogation errors for filter of convolution operation
      Parameters:
      input - input image
      dout - errors from next layer
      outputBlock - output errors
      params - convolution parameters

    • pooling

      public static void pooling(MatrixBlock input, MatrixBlock output, DnnParameters params, LibMatrixDNN.PoolingType poolType)

    • poolingBackward

      public static void poolingBackward(MatrixBlock input, MatrixBlock dout, MatrixBlock outputBlock, DnnParameters params, boolean performReluBackward, LibMatrixDNN.PoolingType poolType)
      This method computes the backpropogation errors for previous layer of pooling operation
      Parameters:
      input - input matrix
      dout - dout matrix
      outputBlock - output matrix
      params - convolution parameters
      performReluBackward - perform ReLU backward
      poolType - type of pooling

    • reluBackward

      public static void reluBackward(MatrixBlock input, MatrixBlock dout, MatrixBlock outputBlock, int numThreads)
      This method computes the backpropagation errors for previous layer of relu operation
      Parameters:
      input - input matrix
      dout - errors from next layer
      outputBlock - output matrix
      numThreads - number of threads

    • biasAdd

      public static void biasAdd(MatrixBlock input, MatrixBlock bias, MatrixBlock outputBlock, int numThreads)
      Performs the operation corresponding to the DML script: ones = matrix(1, rows=1, cols=Hout*Wout) output = input + matrix(bias %*% ones, rows=1, cols=F*Hout*Wout) This operation is often followed by conv2d and hence we have introduced bias_add(input, bias) built-in function
      Parameters:
      input - input matrix
      bias - bias matrix
      outputBlock - output matrix
      numThreads - number of threads

    • channelSums

      public static void channelSums(MatrixBlock input, MatrixBlock outputBlock, int C, int HW)
      Perform channel sum operation
      Parameters:
      input - input matrix block
      outputBlock - output matrix block
      C - number of channels
      HW - height X width

    • batchNorm2DBackward

      public static void batchNorm2DBackward(MatrixBlock image, MatrixBlock dout, MatrixBlock scale, double epsilon, MatrixBlock resultSaveMean, MatrixBlock resultSaveInvVariance, MatrixBlock dX, MatrixBlock dScale, MatrixBlock dBias)

    • batchNorm2D

      public static void batchNorm2D(MatrixBlock image, MatrixBlock scale, MatrixBlock bias, MatrixBlock runningMean, MatrixBlock runningVar, String phase, double epsilon, double mu, MatrixBlock ret, MatrixBlock retRunningMean, MatrixBlock retRunningVar, MatrixBlock resultSaveMean, MatrixBlock resultSaveInvVariance)

    • addBias

      public static void addBias(double[] a, double[] bias, double biasMultiplier, int N, int K, int PQ)

    • multBias

      public static void multBias(double[] a, double[] bias, int N, int K, int PQ)

    • biasMultiply

      public static void biasMultiply(MatrixBlock input, MatrixBlock bias, MatrixBlock outputBlock, int numThreads)
      Performs the operation corresponding to the DML script: ones = matrix(1, rows=1, cols=Hout*Wout) output = input * matrix(bias %*% ones, rows=1, cols=F*Hout*Wout) This operation is often followed by conv2d and hence we have introduced bias_multiply(input, bias) built-in function
      Parameters:
      input - input matrix
      bias - bias matrix
      outputBlock - output matrix
      numThreads - number of threads

    • lstm

      public static void lstm(DnnParameters params)

    • lstmBackward

      public static void lstmBackward(DnnParameters params)