Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.naic.edu/~tghosh/software/wpolyfit.pro Дата изменения: Sun Aug 15 22:19:02 2004 Дата индексирования: Sun Apr 10 07:31:42 2016 Кодировка: Поисковые слова: saturn

;Viewing contents of file
;'/net/www/deutsch/idl/idllib/iuedac/iuelib/pro/wpolyfit.pro'
;
;******************************************************************************
;+
;*NAME:
;
; WPOLYFIT (General IDL Library 01) 7-25-84
;
;*CLASS:
;
; Curve fitting
;
;*CATEGORY:
;
;*PURPOSE:
;
; To fit a weighted polynomial to data points using the
; method of least-squares.
;
;*CALLING SEQUENCE:
;
; WPOLYFIT,X,Y,WEIGHT,NDEG,A,YFIT,CHISQR
;
;*PARAMETERS:
;
; X (REQ) (I) (1) (I L F D)
; Required input vector containing the independent variable
;
; Y (REQ) (I) (1) (F)
; Required input vector containing the dependent variable.
;
; WEIGHT (REQ) (I) (0 1) (F)
; Required input scalar (which is converted to vector) or
; vector giving the weights for each data point.
;
; NDEG (REQ) (I) (0) (I)
; Required input scalar denoting the degree of the polynomial
; to be fit.
;
; A (REQ) (O) (1) (F)
; Required output vector containing the coefficients of the
; polynomial to be fit.
; YFIT=A(0)+A(1)*X+A(2)*X^2+...A(NDEG)*X^NDEG
;
; YFIT (REQ) (O) (1) (F)
; Required output vector containing the calculated values, based
; on the preceding equation, at each value of X.
;
; CHISQ (REQ) (O) (0) (F)
; Required output scalar denoting the reduced chi square statistic
; for the fit.
;
;*EXAMPLES:
;
; To fit a quadratic function to spectral continuum data with uniform
; weighting:
;
; wpolyfit,wcon,fcon,wcon*0.+1.,2,a,yfit,chisq
;
;*SYSTEM VARIABLES USED:
;
; none
;
;*INTERACTIVE INPUT:
;
; none
;
;*SUBROUTINES CALLED:
;
; DETERM
; PARCHECK
; PCHECK
;
;*FILES USED:
;
;*SIDE EFFECTS:
;
;*RESTRICTIONS:
;
;*NOTES:
;
; tested with IDL Version 2.0.10 (sunos sparc) 27 Jun 91
; tested with IDL Version 2.1.0 (ultrix mipsel) 27 Jun 91
; tested with IDL Version 2.1.0 (vms vax) 27 Jun 91
;
;*PROCEDURE:
;
; WPOLYFIT is an IDL version of Bevingtons program POLFIT (p. 140)
; As explained in Bevington, the method of least-squares is used to
; calculate the coefficients. The routine DETERM is used to calculate
; the determinant of the constructed matrices.
;
;*MODIFICATION HISTORY:
;
; ???? I.D. Ahmad initial program
; Jul 25, 1984 RWT GSFC corrected CHISQ calculation for NDEG=0
; and updated documentation
; Jan 30, 1985 RWT GSFC added compilation of PCHECK, allow
; scalar WEIGHT and floating point NDEG.
; Jul 10, 1985 RWT GSFC correct chisq calculation for NDEG>0,
; remove scaling, and add double precision.
; Sep 23, 1985 RWT GSFC modified for DIDL (use double
; precision variables, use N_ELEMENTS,
; remove scaling, and correct CHISQ calculation
; for NDEG>0.
; Apr 15, 1987 RWT GSFC add PARCHECK
; Dec 3, 1987 RWT GSFC add procedure call listing and correct
; error when WEIGHT(0)=0.
; Mar 21, 1988 CAG GSFC add VAX RDAF-style prolog.
; Apr 21, 1988 RWT GSFC use new DETERM_PDP
; Aug 29, 1989 RWT modify for SUN IDL
; Jun 21 1991 GRA CASA cleaned up; tested on SUN, DEC, VAX;
; updated prolog;
; 11 May 94 PJL print a warning if any of the weights are negative
;
;-
;******************************************************************************
pro wpolyfit,xin,yin,weight,ndeg,a,yfit,chisqr
;
npar = n_params(0)
if (npar eq 0) then begin
print,'WPOLYFIT,XIN,YIN,WEIGHT,NDEG,A,YFIT,CHISQR'
retall
endif ; npar eq 0
parcheck,npar,7,'WPOLYFIT'
pcheck,xin,1,010,0011
pcheck,yin,2,010,0011
pcheck,weight,3,110,0011
;
; print a warning if any of the weights are negative
;
temp = where(weight lt 0,count)
if (count gt 0) then begin
print,' '
print,'WARNING: ' + strtrim(count,2) + ' of the ' + $
strtrim(n_elements(weight),2) + ' weight values are negative.'
print,'ACTION: continuing'
print,' '
endif ; count gt 0
;
ndeg = fix(ndeg)
x = double(xin)
y = double(yin)
s = fix(n_elements(x))
chisq = 0 & flag = 0 ; indicates whether x has been adjusted to avoid overflow
nterms = ndeg + 1 ; number of terms in polynomial
array = dblarr(nterms,nterms) ; matrix for solving simultaneous eqns.
a = dblarr(nterms)
w = size(weight)
if (w(0) lt 1) then weight = fltarr(s) + 1.0
;
; if ndeg = 0, simply average y
;
if (ndeg eq 0) then begin
tw = total(weight)
a(0)= total(y*weight) / tw
a0 = a(0)
yfit = 0.*x + a0
chisqr = (yfit - y)
chisqr = total(weight*chisqr*chisqr) / (s - 1)
endif else begin
;
; else proceed with polynomial fit
;
;
; reduce x by an order of magnitude to avoid overflow
;
while 100000000.^(1. / ndeg) le max(x) do begin
x = x / 10.
flag = flag + 1
endwhile
;
; accumulate sums
;
nmax = 2*nterms - 1
sumx = dblarr(nmax)
sumy = a
indx = indgen(nmax)
indy = indgen(n_elements(sumy))
;
; accumulate x^n and y*x^n
;
for i = 0,s-1 do begin
addx = 0.*sumx
addy = 0.*sumy
if (x(i) ne 0.0) then begin
addx = weight(i)*abs(x(i))^indx
addy = weight(i)*y(i)*abs(x(i))^indy
if (x(i) lt 0.) then begin
addx = addx - 2.*(addx)*(indx mod 2)
addy = addy - 2.*(addy)*(indy mod 2)
endif ; x(i) lt 0
endif ; x(i) ne 0
sumx = sumx + addx
sumy = sumy + addy
endfor ; i
chisq = chisq + total(weight*y*y)
;
; construct matrices & calculate coefficients
;
for k = 0,ndeg do $
for j = 0,ndeg do array(j,k) = sumx(j + k)
; determ,array,delta
delta=determ(array)
;
; if matrix is singular, end program
;
if (delta eq 0) then begin
chisqr = 0
a = a * 0
end else begin
for l = 0,ndeg do begin
arr = array
arr(0,l) = sumy(0:ndeg)
; for j = 0,ndeg do arr(j,l) = sumy(j)
;determ,arr,det
det=determ(arr)
a(l) = det / delta
endfor ; l
;
; calculate reduced chi square
;
for j = 0,ndeg do begin
chisq = chisq - 2.*a(j)*sumy(j)
for k = 0,ndeg do chisq = chisq + a(j)*a(k)*sumx(j + k)
endfor ; j
chisqr = float(chisq) / (s - nterms)
;
; calculate yfit
;
yfit = fltarr(s)
xn = yfit + 1.
yfit = yfit + a(0)
for i = 1,ndeg do begin
xn = xn*x
yfit = yfit + a(i)*xn
endfor ; i loop
;
if (flag ne 0) then begin
x = x * 10.0^flag
a = a / (10.0^flag)^indgen(n_elements(a))
endif ; flag ne 0
endelse ; delta ne 0
end ; ndeg ne 0
;
return
end ; wpolyfit