R help - Aug 2012 - All possible models with nls()

Hi all,

I am trying to make a script that prints all possible models from a model
I've created using nls(). It is a logisitc model which in total includes 13
variables. So its >8000 models I need to create, which I don't want to do
manually. I've tried modify scripts made for linear models with no results.

I've tried these scripts on a two variable model (c,a1 and a2 is what I want
to estimate here. Where c is a constant and a1 and a2 is a weight for x1 and
x2):

x1<-sample(0:1,50,replace=T)#explanatory var
x2<-sample(0:1,50,replace=T)#explanatory var
y<-sample(0:1,50,replace=T)#response
# first script I've tried
mod <- list()
for (i in 1:3) {
 	  mod[[i]] <- nls(y~exp(c+(a1*x1)+(a2*x2)
)/
(1+exp(c+(a1*x1)+(a2*x2)))
	,start=list(a1=-0.2,a2=-0.2,c=0.2))
 	  print(summary(mod[[i]]))
 	  }

#second script I've tried
regressors<-c("x1","x2")
regMat<-expand.grid(c(T,F),c(T,F))
regMat

allModelsList <- apply(regMat,1,
function(x)as.formula(nls(y~exp(c+(a1*x1)+(a2*x2)
)/
(1+exp(c+(a1*x1)+(a2*x2)))
	,start=list(a1=-0.2,a2=-0.2,c=0.2))))

allModelsList
allModelsResults <- lapply(allModelsList,
                           function(x) nls(x,
start=list(a1=-0.2,a2=-0.2,c=0.2)))
allModelsResults

They both give me the same models repeatedly, which is not what I wanted.
Any suggestions on how to do this? Please help.
Thanks!
Anna




--
View this message in context:
http://r.789695.n4.nabble.com/All-possible-models-with-nls-tp4641105.html
Sent from the R help mailing list archive at Nabble.com.

Try this.

# the data
y <- sample(0:1, 50, replace=T) # response 
x1 <- sample(0:1, 50, replace=T) # explanatory var 
x2 <- sample(0:1, 50, replace=T) # explanatory var 

# the number of regressors
nregressors <- 2
# all possible combinations
comb <- as.matrix(expand.grid(rep(list(0:1), times=nregressors)))
dimnames(comb)[[2]] <- paste0("x", 1:nregressors)
# create empty list for nls fits
allModelsList <- vector("list", dim(comb)[1])

# fit the nls models
for(i in seq(allModelsList)) {
        # the selected regressors
        sel.regressors <- (1:nregressors)[comb[i, ]==1]
        if(length(sel.regressors) > 0) {
                # create the formula
                snippet <- paste0("(a", sel.regressors, 
                        " * x", sel.regressors, ")",
collapse=" + ")
                my.formula <- paste0("y ~ exp(c + ", snippet, 
                        ")/ (1 + exp(c + ", snippet, "))")
                # create the list of starting values
                my.start <- as.list(c(0.2, rep(-0.2, 
length(sel.regressors))))
                names(my.start) <- c("c", paste0("a",
sel.regressors))
                # fit the nls model
                allModelsList[[i]] <- nls(my.formula, start=my.start)
                } else {
                # if no regressors were selected
                allModelsList[[i]] <- nls(y ~ exp(c) / (1 + exp(c)), 
                        start=list(c=0.2))
                }
        } 

# look at the results
allModelsList

Jean


mael <henriksson.anna@live.se> wrote on 08/23/2012 10:19:38
AM:
> 
> Hi all,
> 
> I am trying to make a script that prints all possible models from a 
model
> I've created using nls(). It is a logisitc model which in total
includes
13
> variables. So its >8000 models I need to create, which I don't want
to
do
> manually. I've tried modify scripts made for linear models with no 
results.
> 
> I've tried these scripts on a two variable model (c,a1 and a2 is what I
want
> to estimate here. Where c is a constant and a1 and a2 is a weight for x1 
and
> x2):
> 
> x1<-sample(0:1,50,replace=T)#explanatory var
> x2<-sample(0:1,50,replace=T)#explanatory var
> y<-sample(0:1,50,replace=T)#response
> # first script I've tried
> mod <- list()
> for (i in 1:3) {
>       mod[[i]] <- nls(y~exp(c+(a1*x1)+(a2*x2)
> )/
> (1+exp(c+(a1*x1)+(a2*x2)))
>    ,start=list(a1=-0.2,a2=-0.2,c=0.2))
>       print(summary(mod[[i]]))
>       }
> 
> #second script I've tried
> regressors<-c("x1","x2")
> regMat<-expand.grid(c(T,F),c(T,F))
> regMat
> 
> allModelsList <- apply(regMat,1,
> function(x)as.formula(nls(y~exp(c+(a1*x1)+(a2*x2)
> )/
> (1+exp(c+(a1*x1)+(a2*x2)))
>    ,start=list(a1=-0.2,a2=-0.2,c=0.2))))
> 
> allModelsList
> allModelsResults <- lapply(allModelsList,
>                            function(x) nls(x,
> start=list(a1=-0.2,a2=-0.2,c=0.2)))
> allModelsResults
> 
> They both give me the same models repeatedly, which is not what I 
wanted.
> Any suggestions on how to do this? Please help.
> Thanks!
> Anna
	[[alternative HTML version deleted]]