However, I’ve noticed that your code does not distinguish between directed and undirected graphs and does not produce eij =eji (it seems that the program always considers first node in the edge list as a source and the second node as a target).

This part should fix it (and it also should run bit faster)

for (i in 1:numatts)
for (j in 1:numatts)
{
iNode <- V(mygraph)[attrib == attlist[i]]
jNode <- V(mygraph)[attrib == attlist[j]]

if (is.directed(mygraph))
mm[i, j] %jNode])
else
mm[i, j] <- length(E(mygraph)[iNode%--%jNode])
}

Thanks for posting the plots–very interesting results.

Have you followed up on the most recent HHP release? In an effort to anonymize data, the Drug/Lab data simply contains information on how many drug prescriptions/lab tests a patient has received in the past year, with no insight as to what the drugs or lab tests are. Not exactly what I was hoping….

I don’t have much experience with panel data, so I’m sure if it’s better to use 3-dimensional arrays, lists of data frames, or what. Perhaps someone else can chime in. I’d also check out the plyr package: http://www.jstatsoft.org/v40/i01/paper

A for loop might be unavoidable for generating iterative data. I’ve done loops like this in matlab and was able to speed them up by an order of magnitude by re-writing it in c, but I’m not sure about R.

As for updating the data within each iteration, I’d do something like this:


df1 <- cohortD(400)
df2 <- df1
df2$age <- df2$age + 1/12
update.index < - runif(nrow(df2) > 0.8
update.values <- sample(df1$fav.flavor, size=sum(update.index), replace=TRUE)
df2$fav.flavor[update.index] <- update.values

# comparing the fav.flavor distributions
x<-melt.data.frame(data.frame(id=df1$id,df1=df1$fav.flavor, df2=df2$fav.flavor),
id.vars="id")
ggplot(x, aes(x=value))+geom_bar(aes(fill=variable), position="dodge")


1) Age each member of the cohort by 1 month
2) With 20% probability, select a new favorite flavor based on the prevalence of favorites in the wider population

Essentially the form of the output of the function would be the same as the data frame, and would just update the entries as specified in the function. Is this possible with the apply approach?

You might also look into the data.table library. It is designed to speed up certain operations on data frames.

I second Adam’s post; this is classic inefficient R programming, and Adam’s method is much better.

What operation are you looking to try to eliminate looping through rows?

I realize it isn’t clear, but I meant the “Cohort C” method, for which I include the (poorly formatted) benchmark results above. Not an implementation in c/Rcpp, which I haven’t tried and seems like overkill for this task.

Adam