Skip to content

randomForest

9 messages · Uwe Ligges, Anirudh Kondaveeti, Liaw, Andy

#
Anirudh Kondaveeti wrote:
See ?randomForest and the argument sampsize.

Uwe Ligges
#
Anirudh Kondaveeti wrote:
No.

Uwe Ligges
#
Uwe had been right all along.  I don't understand what you don't
understand from the documentation.

You can use sampsize=c(300, 300) and replace=FALSE to make sure that all
300 class 1 rows are used, but be warned that that leaves no rows for
OOB estimate.

Andy 

From: Anirudh Kondaveeti
Notice:  This e-mail message, together with any attachme...{{dropped:12}}
#
Anirudh Kondaveeti wrote:
Ah, in that case (stratified sampling) combine arguments "strata" and 
"sampsize", in principle, but you cannot select ALL rows of one class: 
you somehow ignore one of the main ideas of randomForests to bootstrap 
observations - and randomForest will certainly bootstrap for you.

Uwe Ligges
#
Uwe Ligges wrote:
In fact, you can also use  replace = FALSE  as well, but then, as I 
said, one of the main  ideas of randomForest is ignored....

Uwe Ligges