[R] optim - Self-Start values - growth function

[agent dunham]

Dear community, 

I'm trying to model growth with this  function: Yi = A* exp(-k*(1/ti^m)) ; A
asymptote, k rate of decrease of the relative growth rate, m shape
parameter. 

I don't have variable time so, finally,  following some papers, I try to fit 
Yi+a = A*exp(-k* (1/(-k/(log(Yi/A)))^(1/m)+a)^m); 
a= 10

I've tried:

a) nls.1 <- nls(Yi+a ~A*exp(-k* (1/(-k/(log(Yi/A)))^(1/m)+10)^m), data=pgm,
start= list(A=30.3656 , k= 3271.703374, m= -1.870935))    ; start values
obtained: A from data, k & m substituting.  But I obtain this error:

/Error en numericDeriv(form[[3L]], names(ind), env) : 
  Missing value or an infinity produced when evaluating the model /

b) est.o4 <- optim( c(30.3656, 2.841345, 3270.109), funcL, method="SANN",
hessian=TRUE, yr=  data$Yia, x=data$Yi)

where funcL <- function( par, yr, x ) {
A = par[1]
m = par[2]
k = par[3]
y <-  A*exp(-k* (1/(-k/(log(x/A)))^(1/m)+a)^m)
sum((yr-y)^2)   }

Output is as follows: 

/$par
[1] 46.9067228 -0.7824855 16.5397317

$value
[1] 977.2446

$counts
function gradient 
   10000       NA 

$convergence
[1] 0

$message
NULL

$hessian
            [,1]       [,2]       [,3]
[1,]    13.87311  -1583.814  -20.82244
[2,] -1583.81444 190698.126 2526.48636
[3,]   -20.82244   2526.486   33.81710/
/
> eigen(round(est.o4$hessian,4), symmetric= TRUE)
$values
[1] 1.907448e+05 7.786493e-01 2.848952e-01/


I've read here many entries with nls problems, but I'm new with nls and I
don't understand. Why nls fails with my function? Is it nls or my function? 
On the other hand is there any SelfStart function where I can find initial
values for A, k, m parameters?

This is the data I'm dealing with: 
http://r.789695.n4.nabble.com/file/n4257580/data data  Excel Format:  
http://r.789695.n4.nabble.com/file/n4257580/data.xls data.xls 

Any help would be much appreciated, user at host.com











--
View this message in context: http://r.789695.n4.nabble.com/optim-Self-Start-values-growth-function-tp4257580p4257580.html
Sent from the R help mailing list archive at Nabble.com.