[R] Spatial: number of independent components?

[Daniel Malter]

I was missing the spam library. 

I did some testing with m x m matrices (see below). Computing 'a' is the
villain. The computation time for 'a' is exponential in m. For a 100 by 100
matrix, the predicted time is about 20 seconds. Thus, 100,000 runs, would
take about 23 days.

library(igraph)
library(spdep)
library(spam)

d=5:40

tim1=NULL
tim2=NULL
tim3=NULL

for(i in 1:length(d)){
x=rbinom(d[i]^2,1,0.5)
dim(x)=c(d[i],d[i])
tim1[i]=system.time(
        a <- as.spam.listw(nb2listw(cell2nb(nrow(x),ncol(x),torus=T),
style="B"))
        )[3]
tim2[i]=system.time(
        g <- delete.vertices(graph.adjacency(a), which(x!=1)-1)
        )[3]
tim3[i]=system.time(
        clusters(g)
        )[3]
}

reg=lm(log(tim1)~log(d))

par(mfcol=c(1,2))
plot(log(tim1)~log(d))
lines(x=log(d),y=coef(reg)[1]+coef(reg)[2]*log(d),type="l")

plot(tim1~d)
lines(x=d,y=exp(coef(reg)[1]+coef(reg)[2]*log(d)),type="l")


#estimated time for a 100 by 100 matrix
exp(predict(reg,newdata=data.frame(d=100)))

#observed time for a 100 by 100 matrix
d=100
x=rbinom(d^2,1,0.5)
dim(x)=c(d,d)
system.time(a <- as.spam.listw(nb2listw(cell2nb(nrow(x),ncol(x),torus=T),
style="B")))


Best,
Daniel



-- 
View this message in context: http://r.789695.n4.nabble.com/Spatial-number-of-independent-components-tp2262018p2263168.html
Sent from the R help mailing list archive at Nabble.com.