How to detect cycles in a vector in R

Question

Let's say I have the following vector:

vec = c(29, 30, 15, 29, 17, 25, 24, 28, 25, 24, 28, 25, 24, 28, 25, 24, 28)

You'll notice there are three repeating elements (25, 24, and 28). How can I get R to recognize when there are repeating elements (or cycles) in a vector? I want to detect this no matter how many elements are repeating (2 or 5 rather than 3) and no matter how many elements into the vector it starts.

For context, I've got an algorithm that is trying to converge on a value, but sometimes it gets stuck in this repeating loop. I want R to detect when it's stuck in this infinite loop and get out. The vec in my example is a log of the value at each iteration.

I've figured out how I can catch double repeating elements (saving the value from the last iteration to compare to the current iteration) but this 3+ repeating elements has me puzzled.

"cycles" is probably a better description than "repeating elements" — sirallen, Dec 20 '16 at 00:56
Relevant maybe - http://stackoverflow.com/questions/15328925/number-sequence-recognition — thelatemail, Dec 20 '16 at 00:59
Is the vector obtained by function iteration? If so -- you could perhaps use Floyd's cycle finding algorithm (https://en.wikipedia.org/wiki/Cycle_detection ) — John Coleman, Dec 20 '16 at 01:59

score 1 · Answer 1 · answered Dec 20 '16 at 01:19

This function will look for patterns of 2 repeated. I calculate a hash of pairs of element [i] with [i+1] by multiplying the second one by "100" and adding to the first one. You can change this factor to some other number, assuming your integers are bounded by that factor. You might want to change this to 1000000. If you have large integers, you may want to rethink this.

Then I look to make sure the hashes are all unique, i.e. the transition from [i] to [i+1] only happens once.

hasCycle <- function(v) {
  hash <- v[1:length(v)-1] + 100 * v[2:length(v)]
  length(unique(hash)) != length(hash)
}

Here's my test

> a <- c(1, 2,3,4,5,1,6,7)
> hasCycle(a)
[1] FALSE
> 
> b <- c(1, 2,3,4,5,9,7,3,4)
> hasCycle(b)
[1] TRUE

score 0 · Answer 2 · answered Dec 20 '16 at 00:51

0

Before you run your analysis, use the duplicated() method. If the length of the return vector is 0, then there are no duplicates.

answered Dec 20 '16 at 00:51

Mark Lakata

19,989
5
106
123

Good idea. I don't want it to catch values that happen to repeat even if they aren't part of a duplicated pattern, though. I've updated `vec` above. I don't want `29` to get flagged. – CephBirk Dec 20 '16 at 00:55
How do you know that `29` is not a pattern but `25, 24,28` is a pattern? Are you saying that a pattern must have 2 or more repeated elements? If so, then you just need to look for 2 repeated elements. – Mark Lakata Dec 20 '16 at 01:04
I'm interested in finding repeating elements (cycles) greater than length = 1. So in the first case, `29` is followed by `30` while in the second case, it's followed by a `17`. So `29` in this case isn't a part of a cycle. – CephBirk Dec 20 '16 at 01:08

score 0 · Answer 3 · answered Dec 20 '16 at 01:03

This could work:

If I allow vec to run a bit longer:

vec = c(29, 30, 15, 29, 17, 25, 24, 28, 25, 24, 28, 25, 24, 28, 25, 24, 28, 25, 24, 28, 25, 24, 28, 25, 24, 28, 25, 24, 28, 25, 24, 28)

Then I can find cycles up to 10 elements long with this. Longer cycles can be incorporated by changing 10 but I hopefully will never have to deal with that!

any(sapply(1:10, function(i) all(tail(diff(vec, lag = i), 10) == 0)))

How to detect cycles in a vector in R

3 Answers3