Hi Imk,

I think iterator and for-comprehension may help here. I wrote a snippet that implements your first 2 requirements:

    def distance(a: (Double, Double), b: (Double, Double)): Double = ???

    // Defines some total ordering among locations.
    def lessThan(a: (Double, Double), b: (Double, Double)): Boolean = ???

    sc.textFile("input")
      .map { line =>
        val Array(_, latitude, longitude, ip, _, _) = line.split(",")
        ip -> (latitude.toDouble, longitude.toDouble)
      }
      .groupByKey()
      .mapValues { positions =>
        for {
          a <- positions.iterator
          b <- positions.iterator
          if lessThan(a, b) && distance(a, b) < 100
        } yield {
          (a, b)
        }
      }

The key point is that iterators are lazy evaluated, so that you don’t need to store the whole cartesian product.

I didn’t quite get your 3rd requirement, but I think you can implement that following similar approach.

Cheng



On Thu, Jun 5, 2014 at 1:11 PM, lmk <lakshmi.muralikrishnan@gmail.com> wrote:
Hi Oleg/Andrew,
Thanks much for the prompt response.

We expect thousands of lat/lon pairs for each IP address. And that is my
concern with the Cartesian product approach.
Currently for a small sample of this data (5000 rows) I am grouping by IP
address and then computing the distance between lat/lon coordinates using
array manipulation techniques.
But I understand this approach is not right when the data volume goes up.
My code is as follows:

val dataset:RDD[String] = sc.textFile("x.csv")
val data = dataset.map(l=>l.split(","))
val grpData = data.map(r =>
(r(3),((r(1).toDouble),r(2).toDouble))).groupByKey()

Now, I have the data grouped by ipaddress as Array[(String,
Iterable[(Double, Double)])]
ex..
 Array((ip1,ArrayBuffer((lat1,lon1), (lat2,lon2), (lat3,lon3)))

Now I have to find the distance between (lat1,lon1) and (lat2,lon2) and then
between (lat1,lon1) and (lat3,lon3) and so on for all combinations.

This is where I get stuck. Please guide me on this.

Thanks Again.




--
View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/Can-this-be-handled-in-map-reduce-using-RDDs-tp6905p7016.html
Sent from the Apache Spark User List mailing list archive at Nabble.com.