levmar(b,x,y [,f],param [,resid:residmac,deriv:deriv,crit:crvec,\
active:active,maxit:itmax,minit:itmin,print:T])
b REAL vector of starting values for coefficients
x REAL variable, usually a vector or matrix with nrows(x) =
nrows(y)
y REAL vector of data to be fit
f macro called as fit <- f(b,x,param)
param vector or structure of additional parameters for f or NULL
residmac A macro called as residmac(b, x, y, param) to compute a vector
of residuals of length nrows(y). When f is an argument,
'resid' should not be used and residmac(b,x,y,param) is
essentially y - f(b,x,param)
crvect vector(numsig, nsigsq, delta), 3 criteria for convergence
deriv optional macro; deriv(b,x,y,param,j) computes derivative of
f with respect to b[j]
active LOGICAL vector the same length as b
itmax maximum number of iterations permitted (default 30)
itmin minimum number of iterations performed (default 1)
print if T, partial results are printed on each iteration
returns structure(coefs,hessian,jacobian,gradient,rss,residuals,
nobs, iter, iconv)
|
Proper help has not been written yet. Here are comments from the
macro file.
Macro for fitting functions or estimating parameters by minimizing
sums of squares of residuals.
Usage:
levmar(b,x,y [,f],param [,resid:residmac,deriv:deriv,crit:crvec,\
active:active,maxit:itmax,minit:itmin,print:T])
b REAL vector of starting values for coefficients
x REAL variable, usually a vector or matrix with nrows(x) =
nrows(y)
y REAL vector of data to be fit
f macro called as fit <- f(b,x,param)
param vector or structure of additional parameters for f or NULL
residmac A macro called as residmac(b, x, y, param) to compute a vector
of residuals of length nrows(y). When f is an argument,
'resid' should not be used and residmac(b,x,y,param) is
essentially y - f(b,x,param)
crvect vector(numsig, nsigsq, delta), 3 criteria for convergence
controlling the desired accuracy of coefficients and of the
the residual sum of squares, and a target size for the norm
of the gradient.
deriv optional macro; deriv(b,x,y,param,j) computes derivative of
f with respect to b[j]
active LOGICAL vector the same length as b. b[j] "participates"
in the iteration only if active[j] is True (default is
rep(T,length(b)))
itmax maximum number of iterations permitted (default 30)
itmin minimum number of iterations performed (default 1)
print if T, partial results are printed on each iteration
levmar returns structure(coefs,hessian,jacobian,gradient,rss,
residuals,nobs,iter,iconv)
For information on the algorithm, see Brown,K.,M. and Dennis,J.,E.,
Derivative free analogues of the Levenberg-Marquardt and Gauss
algorithms for nonlinear least squares approximation, Numerische
Mathematik, Vol. 18, pp. 289-297 (1972), and Brown,K.,M., Computer
oriented methods for fitting tabular data in the linear and nonlinear
least squares sense, Technical Report No. 72-13, University of
Minnesota Department of Computer and Information Sciences.