Fw : glsm.krige in geoRglm

De: Lho Lho <manguefleur at yahoo.fr>
Objet: glsm.krige in geoRglm
?: "Paulo J Ribeiro Jr" <paulojus at c3sl.ufpr.br>
Date: Mercredi 16 juin 2010, 11h22
Dear Paulo Ribeiro 

Concerning the function krige.glsm(), I copied and past the
example code you have in geoRglm package for binomial data
(b50). 

The following code (from the package) is:

test2 <- glsm.krige(outmcmc.5,
locations=matrix(c(0.15,0.15,0.005,0.05),2,2)) 

My question is:

Why when I try to put a rectangular grid in the locations
argument above (instead of your matrix) 

rectangleb50.grid<-pred_grid(b50$coords[,1:2], by=0.3)

test2 <- glsm.krige(outmcmc.5,
locations=rectangleb50.grid)

I have these 50 warnings() from R:

"In asympvar(kpl.result$predict, messages = FALSE) :
? value of argument lag.max is not suffiently long"

Actually, I am working with counts data and I can perform
glsm.krige only with locations=geodata$coords[,1:2].

If I try to put my rectangular.grid (which I built
according to my data) in the locations argument I have the
same warning messages.

My purpose is to display a image() using the glsm.krige
output

Please, would you like to provide me an explanation?

Best regards

lho






? 

--- En date de?: Lun 24.5.10, Lho Lho <manguefleur at yahoo.fr>
a ?crit?:

De: Lho Lho <manguefleur at yahoo.fr>
Objet: Re: Help with proflik() 2
?: "Paulo J Ribeiro Jr" <paulojus at c3sl.ufpr.br>
Date: Lundi 24 mai 2010, 17h18
Hi Paulo!

Thank you for your help? and sorry for the delay of

my

answer I was away from my office.

It is still not working neither (*100) nor (*1000).

However, I tried in the likfit() function to choose
trend="1st" instead of trend= "2nd" and then, the

proflik()

function worked.

I hope that the problem raised here is only due to

me.

Actually, the quadratic trend surface (trend="2nd")

makes

the data "bell" shaped according to the

plot.geodata()

display.
Then, the geostatistical results I could draw from it

could

be of interest if? compared with geoRglm results
(provided I assume that my burrows counts is poisson
distributed with units.m="m2").

I am going to try again with an other set of data (and

I

cross my finger!!)

Please, let me know if you find out a solution!!!

Hocine


? ? 




--- En date de?: Dim 23.5.10, Paulo J Ribeiro Jr

<paulojus at c3sl.ufpr.br>

a ?crit?:

De: Paulo J Ribeiro Jr <paulojus at c3sl.ufpr.br>
Objet: Re: Help with proflik()
?: "Lho Lho" <manguefleur at yahoo.fr>
Date: Dimanche 23 mai 2010, 21h04
Hi 

Yhanks for reporting and sending detailed

information.

I have not yet inspected/run the files and

commands

but I suspect that the probelem may be due to

numerical

instability
due to the different orders of magnitude between

your

phi

(range)
parameters and the others.

As a quick try you could multiply all the

coordinates

by a

constant (100
or 1000), changind range parameters (phi) values
accordingly and see if
the problem persists

Please let me know

best
P.J.


Em Dom, 2010-05-23 ?s 17:37 +0000, Lho Lho

escreveu:

Dear Paulo Justiniano Ribeiro 

After performing the proflik() function I

had

this

message from R:

##proflik:computing profile likelihood for

the

sill

Error in solve.default(xix, xiy) :
???system is computationally singular:

reciprocal condition number=5.05981e-19##

I guess it is related to "matrice" with the

term

solve.default(??)

Unfortunately, I was not able to overcome

this

problem.

Would you like to help me?


Please find attached my data in text format

(Apple).

My? R code was :

ini.values=expand.grid(x=seq(500,1300,by=25),y=seq(0,0.40,by=0.01))

inivalues<-as.matrix(ini.values)

Burrows.likfit<-likfit(Burrows.geo,trend="2nd",fix.lambda=TRUE,lambda=1,ini.cov.pars=ini.values,fix.nugget=FALSE,nugget=0,lik.method="ML",cov.model="mat")

profile<-proflik(Burrows.likfit,geodata=Burrows.geo,sill.values=seq(1210,1213,l=4),range.values=seq(0.026,0.029,l=4),nugget.values=seq(31,34,l=4),

uni.only=T))

Hocine BENCHIKH
PhD student UCC
Ireland




? ? ???

? ?