Hello every one,
I am a PhD student in Statistics, for my work I had to modify the rpart code
and use it to build some decision trees.
I thought I managed, but I noticed some strangeness in the trees I got by
using the modified rpart.
I'd like to ask you if I did the right modification:
In fact in rpart it is the gini measure that I would like to modify:
as far as I know the gini measure is of the form gini(t) =1 -
sum(i=1:n)[Fi(t)] with Fi(t)=Ni(t)/N(t)=p(t)
I wanted to replace this measure by m(t) 1-0.5sum(i=1:n)[Fi(t)log2(Fi(t)+1)]
When I look into the rpart package, in the src, in gini.c, I found
static double gini_impure1(p) double p; { return(1 - p*p); }
which I replaced by
static double gini_impure1(p) double p; { return(1-0.5*p*log2(p+1)); }
am I right?
Thank you
Nicolas
--
View this message in context:
http://r.789695.n4.nabble.com/Modify-rpart-tp3457430p3457430.html
Sent from the R help mailing list archive at Nabble.com.