Classification Trees and basic Random Forest pkg using t ree structures in C

From: Hin-Tak Leung

Izmirlian, Grant (NIH/NCI) wrote:
<snipped>
The only interesting feature is that the tree structure has been
implemented in C. Its a neater way to carry stuff around and I am 
guessing would make future implementation easier.

Because of its inherent redundancy from the users standpoint, it
isn't something to send to CRAN. However, I was wondering whether
anyone is interested in a copy?
Hi,

Hmm, why didn't you just post a URL?
Isn't it a bit too much to assume that everyone has a personal web space
somewhere?
Incidentally I am actually very
interested in seeing your code. I am working on a project where
the data set is extremely large, but the permuntation of the states of
the data is extremely small. Each piece of data consists of only 4 
states, so stuffing it as an R object (which takes up 32-byte? on
32-bit machines) or even an char vector is quite wasteful; so I
have written a "strange" data.frame where internally it uses only
2-bit for storage. (it is still work-in-process but I have got to
the point of being able to get and set each 2-bit cell now).
For some of the data we encounter, all X variables are binary, so each data
point can be encoded into a bitstring.  There are algorithms that take
advantage of that.  The problem is interfacing such code with R.  I know of
no good solutions.  As I told Grant, I thought about what he did, too, but
the difficulty is how to pass such data structures to R.  Actually, some
time down the road I might try to use the dendrogram class that's in R, and
manipulate them in C.  Not sure about efficiency though. 

Andy
Hin-Tak Leung

______________________________________________
R-devel at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel

For some of the data we encounter, all X variables are binary, so each data
point can be encoded into a bitstring.  There are algorithms that take
advantage of that.  The problem is interfacing such code with R.  I know of
no good solutions.  As I told Grant, I thought about what he did, too, but
the difficulty is how to pass such data structures to R.  Actually, some
time down the road I might try to use the dendrogram class that's in R, and
manipulate them in C.
I faced similar problems some time ago and ended up representing a
(binary) tree as recursive lists which can be manipulated from both the C
and R side. The `party' package has the code (and an internal random
forest function, however, without R interface yet) and the vignette
explains some details.

Best,

Torsten
Not sure about efficiency though.

Andy

Hin-Tak Leung

______________________________________________
R-devel at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel

______________________________________________
R-devel at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel

From: Hin-Tak Leung
<snipped>
Hmm, why didn't you just post a URL?
Isn't it a bit too much to assume that everyone has a personal web space
somewhere?
Just for the sake of argument... I did assume that nih.gov is a sizeable
government organization and have official channnels for such things.
That's what government agencies do, and I suppose this software work is
in-the-line-of-duty for public consumption and therefore quite 
appropriate to put on a *.gov web site. (the same applies to *.ac.uk and 
  *.edu postings).
For some of the data we encounter, all X variables are binary, so each data
point can be encoded into a bitstring.  There are algorithms that take
advantage of that.  The problem is interfacing such code with R.  I know of
no good solutions.  As I told Grant, I thought about what he did, too, but
the difficulty is how to pass such data structures to R.  Actually, some
time down the road I might try to use the dendrogram class that's in R, and
manipulate them in C.  Not sure about efficiency though. 
The best examples I have seen of manipulating foreign object is among 
the omegahat projects, like RSPerl and PSPython. Quite insteresting reading.

Hin-Tak Leung