obj.f
Code for FUNMIN  using two-hidden layers sigmoid network to fit a data set.
Analytical gradients are coded as well. (SUBROUTINE GRANAL)
There is the option for weight-decay regularization. The penalty parameter is set by flag 2.
To print results, FLAG 1 must be set to equal 1

MACROF
Two macros: .p   and .in
.in  sets the analytic derivative mode and the penalty strength to 1.d-6
.p   writes: x,y,N(x,w),N(x,w)-y   to file "RES" for plotting

NET_PARAMS,  POINT_PARAMS
Both contain parameter declarations and are used in obj.f  as included files, namely by:
      include 'NET_PARAMS'
      include 'POINT_PARAMS'
	  
oo
contains random numbers uniformly distributed in (-1,1) for initialization purposes.
(Merlin command:  init 1 oo)

TwoLayersNN.pdf
Contains the relevant formulas for a two-hidden layers sigmoid ANN