[RsR] number of subsamples in robustbase::lmrob.S()

Dear All,

I'm trying to estimate the same model
twice using the FastS with a varying
# of ('simple') sub-samples. 

The man page for lmrob.control() defines 
an option 'mts' which seems to be what I 
want:

mts maximum number of samples to try in subsampling algorithm.

However, when trying on a dataset:

library(robustbase)
#my data:
n<-500
p<-10
x0<-matrix(rnorm(n*(p-1)),nc=p-1)
y0<-rnorm(n)

s1<-lmrob.control()
s1$subsampling<-"simple"
s1$max.it<-500
s1$k.max<-500
s1$maxit.scale<-500

#\# subsamples?
s1$mts<-10000
system.time(lmrob.S(x=cbind(1,x0),y=y0,control=s1))
 user  system elapsed 
 0.084   0.000   0.081 
#\# subsamples?
s1$mts<-100
system.time(lmrob.S(x=cbind(1,x0),y=y0,control=s1))
 user  system elapsed 
 0.076   0.000   0.077


I find that the two timing are very close...I suspect 
they both use a similar number of starting p-subsets 
(and that I set the options in lmrob.control() wrongly). 

How can I check how many starting p-subsets were used 
in each estimations?

Best regards,

Dear Kaveh,

mts sets just the number of samples the algorithm tries before giving up finding a nonsingular subsample (if simple subsampling is used).
nResample is the parameter you want.

Best,

Manuel

On 1. May 2013, at 13:11 , Kaveh Vakili <kaveh.vakili at wis.kuleuven.be> wrote:

Dear All,

I'm trying to estimate the same model
twice using the FastS with a varying
# of ('simple') sub-samples.

The man page for lmrob.control() defines
an option 'mts' which seems to be what I
want:

mts maximum number of samples to try in subsampling algorithm.

However, when trying on a dataset:

library(robustbase)
#my data:
n<-500
p<-10
x0<-matrix(rnorm(n*(p-1)),nc=p-1)
y0<-rnorm(n)

s1<-lmrob.control()
s1$subsampling<-"simple"
s1$max.it<-500
s1$k.max<-500
s1$maxit.scale<-500

#\# subsamples?
s1$mts<-10000
system.time(lmrob.S(x=cbind(1,x0),y=y0,control=s1))
  user  system elapsed
  0.084   0.000   0.081
#\# subsamples?
s1$mts<-100
system.time(lmrob.S(x=cbind(1,x0),y=y0,control=s1))
  user  system elapsed
  0.076   0.000   0.077


I find that the two timing are very close...I suspect
they both use a similar number of starting p-subsets
(and that I set the options in lmrob.control() wrongly).

How can I check how many starting p-subsets were used
in each estimations?

Best regards,