This Least Square Estimation demo contains 3 prgrammes.

LSQFT.F   
^^^^^^^
* Input is through 'fit.inp'
  - Record 1  : 3 integers.  
    np- no of parameters in fit
    nx- size of design vector (dimension of x)
    nd- number of experiments (data points) for fit
  - Next nd records contain 'nx+1' floats per record
    nx are the values of each compnent of x
    last is the value of of y (the response)
    Note y=f(X) = f{x(1), x(2), . . . .x(nx) } 
  - Last record : 1 integer
    if 1 = shift origin to average of data
       0 = do not shift origin.

* Shifting of origin if employed when not using 
  coded variables (-1 to +1) , can reduce numerical
  errors

* LSQFT.F requires one subroutine 'setf' to be 
  defined.  This evaluates the X matrix given
  all designs.
  The defaults setf contains:
  X = {1.   x(1)    x(2) }   for a nx = 2 problem

* Output is in 'fit.out'

* Also creates another output file 'repeat'
  This file should exist.  It keeps appending
  to this file.
  Incase you do not have this file do --> 'touch repeat'


DATA.F
^^^^^^
* This program generates 'fit.inp' for LSQFT.F
  yt = 3  + 2* x(1)  +  x(2)  : is the function

* ye = yt + eps;  where  eps = RN(mu=0, sigma=0.2)
  Note you may edit the programme to change
  these defaults.

* Everytime you run this programe, it will prompt
  you for a seed to generate a random number
  By giving different seeds you can influence
  data

* Output file is 'fit.inp' for 20 designs

* You may also use it for 2^k factorial designs


AVERAGE.F
^^^^^^^^^

* Supposing you run 'DATA' and 'LSQFT' in 
  a loop by giving different seeds each time.
  This writes 'b' vector in a file called 'repeat'

* AVERAGE reads in 'repeat' and prints out average
  of all b estimates.