Question

Suppose you have a bunch of computations, which you want to execute using foreach. Perhaps you want to parallelize them, now or in the future. Each computation returns a single number. What is the best idiom to obtain the mean of all the numbers computed in this fashion?

I have two solutions in mind, which I’ll post as possible answers so that you can vote and comment on these individually, and perhaps even edit them to improve them. But I guess there might be even better ways, so additional answers are welcome. My answers will include examples which you can adopt to demonstrate your own techniques.

Answer 1

One solution accumulates the results in a vector, and eventually executes the mean function on that.

> foreach(i = icount(300), .combine = c, .inorder = FALSE, .multicombine = TRUE,
+         .final = mean) %do% { i*7 %% 11 }
[1] 1053.5

Benefits:

• Works with nested foreach calls as well
• Works with aggregating functions other than mean as well
• Does not require any knowledge about the number of values in the iteration

Drawbacks:

• Collects all the results into a vector first, which might consume unneccessary amounts of memory, particularly when performing a huge number of very small jobs.