Thank you Todd, Silvio...

I had to stare at Silvio's answer for a while.

_If I'm interpreting the aggregateByKey() statement__correctly ...
_   (Within-Partition Reduction Step)
   a: is a TUPLE that holds: (runningSum, runningCount).
   b: is a SCALAR that holds the next Value

  (Cross-Partition Reduction Step)
  a: is a TUPLE that holds: (runningSum, runningCount).
  b: is a TUPLE that holds: (nextPartitionsSum, nextPartitionsCount).

Under that interpretation, I tried to write & run the Python equivalent, like so: rdd1.aggregateByKey(0, lambda a,b: (a[0] + b, a[1] + 1), lambda a,b: (a[0] + b[0], a[1] + b[1]))

Sadly, it didn't work, yielding the following exception which indicates that the indexing above is incorrect:
    lambda v: seqFunc(createZero(), v), seqFunc, combFunc, numPartitions)
    File "<input>", line 1, in <lambda>
    TypeError: 'int' object has no attribute '__getitem__'

/Sidenote: Surprisingly, there isn't much documentation -- at least not for Python -- for this useful aggregateByKey()// //method and use case; although I will be sure to write a //g//ist today, once I get this working. :)/
_
__I think I'm nearly there though, so..._
    (1) Is my written interpretation above about of what (a,b) are correct?
    (2) If yes, what then, is getting passed in the Python case?

I guess I'm looking for the Python equivalent to the first statement in Silvio's answer (below). But my
reasoning to deconstruct and reconstruct is missing something.

Thanks again!




On 04/28/2015 11:26 AM, Silvio Fiorito wrote:
If you need to keep the keys, you can use aggregateByKey to calculate an avg of the values:

val step1 = data.aggregateByKey((0.0, 0))((a, b) => (a._1 + b, a._2 + 1), (a, b) => (a._1 + b._1, a._2 + b._2))
val avgByKey = step1.mapValues(i => i._1/i._2)

Essentially, what this is doing is passing an initializer for sum and count, then summing each pair of values and counting the number of values. The last argument is to combine the results of each partition, if the data was spread across partitions. The result is a tuple of sum and count for each key.

Use mapValues to keep your partitioning by keys intact and minimize a full shuffle for downstream keyed operations. It just calculates the avg for each key.

From: Todd Nist
Date: Tuesday, April 28, 2015 at 10:20 AM
To: "subscripti...@prismalytics.io <mailto:subscripti...@prismalytics.io>"
Cc: "user@spark.apache.org <mailto:user@spark.apache.org>"
Subject: Re: Calculating the averages for each KEY in a Pairwise (K,V) RDD ...

Can you simply apply the https://spark.apache.org/docs/1.3.1/api/scala/index.html#org.apache.spark.util.StatCounter to this? You should be able to do something like this:

val stats = RDD.map(x => x._2).stats()

-Todd

On Tue, Apr 28, 2015 at 10:00 AM, subscripti...@prismalytics.io <mailto:subscripti...@prismalytics.io> <subscripti...@prismalytics.io <mailto:subscripti...@prismalytics.io>> wrote:

    Hello Friends:

    I generated a Pair RDD with K/V pairs, like so:

    >>>
    >>> rdd1.take(10) # Show a small sample.
     [(u'2013-10-09', 7.60117302052786),
     (u'2013-10-10', 9.322709163346612),
     (u'2013-10-10', 28.264462809917358),
     (u'2013-10-07', 9.664429530201343),
     (u'2013-10-07', 12.461538461538463),
     (u'2013-10-09', 20.76923076923077),
     (u'2013-10-08', 11.842105263157894),
     (u'2013-10-13', 32.32514177693762),
     (u'2013-10-13', 26.249999999999996),
     (u'2013-10-13', 10.693069306930692)]

    Now from the above RDD, I would like to calculate an average of
    the VALUES for each KEY.
    I can do so as shown here, which does work:
    *
    *>>>**countsByKey = sc.broadcast(rdd1.countByKey()) # SAMPLE
    OUTPUT of countsByKey.value: {u'2013-09-09': 215, u'2013-09-08':
    69, ... snip ...}
    >>> rdd1 = rdd1.reduceByKey(operator.add) # Calculate the
    numerator (i.e. the SUM).
    >>> rdd1 = rdd1.map(lambda x: (x[0],
    x[1]/countsByKey.value[x[0]])) # Divide each SUM by it's
    denominator (i.e. COUNT)
    >>> print(rdd1.collect())
      [(u'2013-10-09', 11.235365503035176),
       (u'2013-10-07', 23.39500642456595),
       ... snip ...
      ]

    But I wonder if the above semantics/approach is the optimal one;
    or whether perhaps there is a single API call
    that handles common use case.

    Improvement thoughts welcome. =:)

    Thank you,
    nmv



--
PRISMALYTICS Sincerely yours,
Team PRISMALYTICS
------------------------------------------------------------------------
PRISMALYTICS, LLC. <http://www.prismalytics.com/> | www.prismalytics.com <http://www.prismalytics.com/> P: 212.882.1276 <tel:212.882.1276> | subscripti...@prismalytics.io <mailto:subscripti...@prismalytics.io> Follow Us: https://www.LinkedIn.com/company/prismalytics <https://www.linkedin.com/company/prismalytics>

Prismalytics, LLC. <http://www.prismalytics.com/>
data analytics to literally count on

Reply via email to