SGDF loop processing time

Dear Roger, and List-Serve,

I am posting here a .RData file(IlonaQuesRgeo.RData) to help with my
previous query (see original email below) as per Roger's request. Please let
me know if the attachment does not work.
Hi,

It reached me, not sure about the list - the posting guide on R-help sets 
out the attachments policy.

library(sp)
load("IlonaQuesRgeo.RData")

Patchmap$HSvalue <- HSmap$HSvalue
Patchmap_orig <- Patchmap
fullgrid(Patchmap) <- FALSE
gridded(Patchmap) <- FALSE
# drop back to points to split on ID

Patchmap0 <- Patchmap[Patchmap$ID==0,]
Patchmap1 <- Patchmap[Patchmap$ID==1,]
# split subsets

Patchmap0$NPHS <- ifelse(Patchmap0$HSvalue < HSthreshold,
   Patchmap0$HSvalue, NA)
# vectorised ID==0 (is the else value correct?)

Patchmap1$NPHS <- round(rnorm(length(Patchmap1$ID), mean_HS,
   sd_HS),digits=3)
Patchmap1$NPHS <- ifelse(Patchmap1$NPHS < new_HSthreshold,
   Patchmap1$HSvalue, Patchmap1$NPHS)
# vectorised ID==1 (is the else value correct?)

Patchmap_res <- rbind(Patchmap0, Patchmap1)
gridded(Patchmap_res) <- TRUE
fullgrid(Patchmap_res) <- TRUE
# put back together, re-sorting to get back in place

all.equal(Patchmap_res$ID, Patchmap_orig$ID)
# sanity check on IDs

Patchmap_orig$NPHS <- Patchmap_res$NPHS
# assign back if need be and display image
image(Patchmap_orig, "NPHS")

There must be better ways, the workflow is a little contorted, but it may 
be worth trying. Maybe a set.seed() to anchor the rnorm()?

Hope this helps,

Roger
Cheers,
Ilona

Ilona Naujokaitis-Lewis
Centre for Applied Conservation Research
Forest Sciences Department
University of British Columbia
3041-2424 Main Mall
Vancouver, BC V6T 1Z4

phone: 604 822.4382
email: ilonan at interchange.ubc.ca

-----Original Message-----
From: Roger Bivand [mailto:Roger.Bivand at nhh.no]
Sent: Wednesday, June 25, 2008 10:18 AM
To: Ilona Naujokaitis-Lewis
Cc: r-sig-geo at stat.math.ethz.ch
Subject: Re: [R-sig-Geo] SGDF loop processing time

On Wed, 25 Jun 2008, Ilona Naujokaitis-Lewis wrote:

Dear List-serve,

I have written a loop that creates a Spatial Grid Dataframe from 2 other
SGDFs. I am interested in finding out if there is a way to decrease the
processing time of the entire loop. I am aware that loops in R are not the
most efficient way to go, but based on my knowledge, I am not sure how to
get around this. To avoid problems with memory allocation I have added
gc()
command throughout the loop.

The system.time() results for the loop (see code below) with a small
(16x12)
grid is:
user  system elapsed
17.08    0.02   17.22

The problem is when I work with larger grids (i.e. greater than
dimensions:
900x1200). The processing time is in the order of 24 hours.

I have checked threads for help on both decreasing the memory size, and
writing efficient code. Any suggestions on how I might decrease the
processing time, as this is the biggest concern right now (both using and
not using R, but the former preferred), would be greatly appreciated.

Here is the code with a brief description of the objects.

Objects:
prePatchmap: SGDF, binary (0,1)
mean_HS: mean habitat suitability value of original habitat suitability
map
sd_HS: standard deviation of original habitat suitability map
new_HSthreshold: new habitat suitability threshold
HSthreshold: original habitat suitability threshold
HSmap: SGDF of original habitat suitability values
Could you provide an online example - save() the objects you need here and
post the RData file? This does look vectorisable, at least in two chunks.

Roger

for (i in 1:length(prePatchmap$patch))
   {
   if (prePatchmap$patch[i]==1)
      {
      prePatchmap$NPHS[i]<-round(rnorm(1, mean_HS, sd_HS),digits=3)
      gc()
         {
         if (prePatchmap$NPHS[i]<new_HSthreshold)
           {
           prePatchmap$NPHS[i]<-new_HSthreshold
           gc()
           }
         }
      } else
      {
      if (HSmap$HSvalue[i]<HSthreshold)
           {
           prePatchmap$NPHS[i]<-HSmap$HSvalue[i]
           }
      gc()
      }
    }

Cheers,

Ilona Naujokaitis-Lewis

Ilona Naujokaitis-Lewis
Centre for Applied Conservation Research
Forest Sciences Department
University of British Columbia
3041-2424 Main Mall
Vancouver, BC V6T 1Z4

phone: 604 822.4382
email: ilonan at interchange.ubc.ca

_______________________________________________
R-sig-Geo mailing list
R-sig-Geo at stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-geo

Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no

SGDF loop processing time

Thread (4 messages)