>Y[,1] [,2] [,3] [1,] 0 1 0 [2,] 0 1 0 [3,] 0 0 1 [4,] 1 0 0 [5,] 0 0 1 [6,] 0 0 1 [7,] 1 0 0 [8,] 1 0 0 [9,] 0 0 1 [10,] 1 0 0>Xpri82 pan82 1 0 0 2 0 0 3 1 0 4 1 0 5 0 1 6 0 0 7 1 0 8 1 0 9 0 0 10 0 0>K=2 >J=3LL <- function(b=rep(0,(J-1)*K)){ B=matrix(c(b,rep(0,K)),ncol=J,nrow=K) fit <- X%*%B p<-exp(fit)/rowSums(exp(fit)) sum(Y*log(p)) } grad<- function(b=rep(0,(J-1)*K)){ B=matrix(c(b,rep(0,K)),ncol=J,nrow=K) fit <- X%*%B p<-exp(fit)/rowSums(exp(fit)) Yp <- Y-p Yp<-matrix(rep(t(Yp),each=K),ncol=K*J,by=T) X <- matrix(rep(X,J) ,ncol=K*J) apply(Yp*X,2,sum) } library(stats4) mle(LL) Error in validObject(.Object) : invalid class "mle" object: invalid object for slot "fullcoef" in class "mle": got class "list", should be or extend class "numeric" mle(LL,gr=grad) Error in optim(start, f, method = method, hessian = TRUE, ...) : gradient in optim evaluated to length 6 not 0 what is wrong with my code?I try to fix it myself but fails,anyone helps me ?Thank you! -- Deparment of Sociology Fudan University
ronggui <ronggui.huang at gmail.com> writes:> >Y > [,1] [,2] [,3] > [1,] 0 1 0 > [2,] 0 1 0 > [3,] 0 0 1 > [4,] 1 0 0 > [5,] 0 0 1 > [6,] 0 0 1 > [7,] 1 0 0 > [8,] 1 0 0 > [9,] 0 0 1 > [10,] 1 0 0 > > >X > pri82 pan82 > 1 0 0 > 2 0 0 > 3 1 0 > 4 1 0 > 5 0 1 > 6 0 0 > 7 1 0 > 8 1 0 > 9 0 0 > 10 0 0 > > >K=2 > >J=3 > > > LL <- function(b=rep(0,(J-1)*K)){ > B=matrix(c(b,rep(0,K)),ncol=J,nrow=K) > fit <- X%*%B > p<-exp(fit)/rowSums(exp(fit)) > sum(Y*log(p)) > } > > grad<- function(b=rep(0,(J-1)*K)){ > B=matrix(c(b,rep(0,K)),ncol=J,nrow=K) > fit <- X%*%B > p<-exp(fit)/rowSums(exp(fit)) > Yp <- Y-p > Yp<-matrix(rep(t(Yp),each=K),ncol=K*J,by=T) > X <- matrix(rep(X,J) ,ncol=K*J) > apply(Yp*X,2,sum) > } > > library(stats4) > mle(LL) > Error in validObject(.Object) : invalid class "mle" object: invalid > object for slot "fullcoef" in class "mle": got class "list", should be > or extend class "numeric"It is a bit strange that it gets dicovered so late (when the mle object is printed), but the issue is that the negative log likelihood has to be a function of a number of *scalar* parameters. Vector parameters are not allowed (yet? this is not the first case where people have expected or desired that they were allowed).> mle(LL,gr=grad) > Error in optim(start, f, method = method, hessian = TRUE, ...) : > gradient in optim evaluated to length 6 not 0 > > what is wrong with my code?I try to fix it myself but fails,anyone > helps me ?Thank you! > > -- > Deparment of Sociology > Fudan University > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html >-- O__ ---- Peter Dalgaard ??ster Farimagsgade 5, Entr.B c/ /'_ --- Dept. of Biostatistics PO Box 2099, 1014 Cph. K (*) \(*) -- University of Copenhagen Denmark Ph: (+45) 35327918 ~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk) FAX: (+45) 35327907
I see two problems:
First, "mle" expects its first argument to return
(-log(likelihood));
your "LL" is closer to "+log(likelihood)", but it seems to
be missing
something beyond that. In particular, "Y" is 10 x 3, but
"fit" is only
10 x 2.
Second, I'm not sure, but it looks to me like "mle" wants to
optimize
over scalar arguments its first argument, "LL" in your case, and you
want it to optimize over a vector. I can't see where the help page for
"mle" spells it out, but unless I missed something, the examples all
used scalar arguments. Moreover, I walked through your call to "mle"
line by line after 'debug(mle)'. The argument "start" that
"mle" passes
to "optim" was of length 0. Therefore, you need to do something
different to convince "mle" that it needs to optimize by varying more
than 0 parameters.
Have you considered changing the sign of the output of "LL" (which
you need to do anway) and giving it directly to "optim"? That should
work.
Moreover, it looks to me like you want to do multinomial logistic
regression. If that is so, have you considered searching for that in
particular? RSiteSearch("multinomial logistic regression") returned
130
hits for me just now.
hope this helps.
spencer graves
ronggui wrote:
>>Y
>
> [,1] [,2] [,3]
> [1,] 0 1 0
> [2,] 0 1 0
> [3,] 0 0 1
> [4,] 1 0 0
> [5,] 0 0 1
> [6,] 0 0 1
> [7,] 1 0 0
> [8,] 1 0 0
> [9,] 0 0 1
> [10,] 1 0 0
>
>
>>X
>
> pri82 pan82
> 1 0 0
> 2 0 0
> 3 1 0
> 4 1 0
> 5 0 1
> 6 0 0
> 7 1 0
> 8 1 0
> 9 0 0
> 10 0 0
>
>
>>K=2
>>J=3
>
>
>
> LL <- function(b=rep(0,(J-1)*K)){
> B=matrix(c(b,rep(0,K)),ncol=J,nrow=K)
> fit <- X%*%B
> p<-exp(fit)/rowSums(exp(fit))
> sum(Y*log(p))
> }
>
> grad<- function(b=rep(0,(J-1)*K)){
> B=matrix(c(b,rep(0,K)),ncol=J,nrow=K)
> fit <- X%*%B
> p<-exp(fit)/rowSums(exp(fit))
> Yp <- Y-p
> Yp<-matrix(rep(t(Yp),each=K),ncol=K*J,by=T)
> X <- matrix(rep(X,J) ,ncol=K*J)
> apply(Yp*X,2,sum)
> }
>
> library(stats4)
> mle(LL)
> Error in validObject(.Object) : invalid class "mle" object:
invalid
> object for slot "fullcoef" in class "mle": got class
"list", should be
> or extend class "numeric"
>
> mle(LL,gr=grad)
> Error in optim(start, f, method = method, hessian = TRUE, ...) :
> gradient in optim evaluated to length 6 not 0
>
> what is wrong with my code?I try to fix it myself but fails,anyone
> helps me ?Thank you!
>
> --
> Deparment of Sociology
> Fudan University
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html