NF01AY

Computing the output of a set of neural networks

Purpose

  To calculate the output of a set of neural networks with the
  structure

          - tanh(w1'*z+b1) -
        /      :             \
      z ---    :           --- sum(ws(i)*...)+ b(n+1)  --- y,
        \      :             /
          - tanh(wn'*z+bn) -

  given the input z and the parameter vectors wi, ws, and b,
  where z, w1, ..., wn are vectors of length NZ, ws is a vector
  of length n, b(1), ..., b(n+1) are scalars, and n is called the
  number of neurons in the hidden layer, or just number of neurons.
  Such a network is used for each L output variables.

      SUBROUTINE NF01AY( NSMP, NZ, L, IPAR, LIPAR, WB, LWB, Z, LDZ,
     $                   Y, LDY, DWORK, LDWORK, INFO )
C     .. Scalar Arguments ..
      INTEGER           INFO, L, LDWORK, LDY, LDZ, LIPAR, LWB, NSMP, NZ
C     .. Array Arguments ..
      DOUBLE PRECISION  DWORK(*), WB(*), Y(LDY,*), Z(LDZ,*)
      INTEGER           IPAR(*)

Input/Output Parameters

  NSMP    (input) INTEGER
          The number of training samples.  NSMP >= 0.

  NZ      (input) INTEGER
          The length of each input sample.  NZ >= 0.

  L       (input) INTEGER
          The length of each output sample.  L >= 0.

  IPAR    (input) INTEGER array, dimension (LIPAR)
          The integer parameters needed.
          IPAR(1) must contain the number of neurons, n, per output
          variable, denoted NN in the sequel.  NN >= 0.

  LIPAR   (input) INTEGER
          The length of the vector IPAR.  LIPAR >= 1.

  WB      (input) DOUBLE PRECISION array, dimension (LWB)
          The leading (NN*(NZ+2)+1)*L part of this array must
          contain the weights and biases of the network. This vector
          is partitioned into L vectors of length NN*(NZ+2)+1,
          WB = [ wb(1), ..., wb(L) ]. Each wb(k), k = 1, ..., L,
          corresponds to one output variable, and has the structure
          wb(k) = [ w1(1), ..., w1(NZ), ..., wn(1), ..., wn(NZ),
                    ws(1), ..., ws(n), b(1), ..., b(n+1) ],
          where wi(j) are the weights of the hidden layer,
          ws(i) are the weights of the linear output layer, and
          b(i) are the biases, as in the scheme above.

  LWB     (input) INTEGER
          The length of the array WB.
          LWB >= ( NN*(NZ + 2) + 1 )*L.

  Z       (input) DOUBLE PRECISION array, dimension (LDZ, NZ)
          The leading NSMP-by-NZ part of this array must contain the
          set of input samples,
          Z = ( Z(1,1),...,Z(1,NZ); ...; Z(NSMP,1),...,Z(NSMP,NZ) ).

  LDZ     INTEGER
          The leading dimension of the array Z.  LDZ >= MAX(1,NSMP).

  Y       (output) DOUBLE PRECISION array, dimension (LDY, L)
          The leading NSMP-by-L part of this array contains the set
          of output samples,
          Y = ( Y(1,1),...,Y(1,L); ...; Y(NSMP,1),...,Y(NSMP,L) ).

  LDY     INTEGER
          The leading dimension of the array Y.  LDY >= MAX(1,NSMP).

  DWORK   DOUBLE PRECISION array, dimension (LDWORK)

  LDWORK  INTEGER
          The length of the array DWORK.  LDWORK >= 2*NN.
          For better performance, LDWORK should be larger.

  INFO    INTEGER
          = 0:  successful exit;
          < 0:  if INFO = -i, the i-th argument had an illegal
                value.

  BLAS routines are used to compute the matrix-vector products.

  None

Program Text

  None

  None

  None