How to refer Spark RDD element multiple times using underscore notations.
For example I need to convert RDD[String] to RDD[(String, Int)]. I can create anonymous function using function variables but I would like to do this using Underscore notation. How I can achieve this.
PFB sample code.
val x = List("apple", "banana")
val rdd1 = sc.parallelize(x)
// Working
val rdd2 = rdd1.map(x => (x, x.length))
// Not working
val rdd3 = rdd1.map((_, _.length))
Why does the last line above not work?
An underscore or (more commonly) a placeholder syntax is a marker of a single input parameter. It's nice to use for simple functions, but can get tricky to get right with two or more.
You can find the definitive answer in the Scala language specification's Placeholder Syntax for Anonymous Functions:
An expression (of syntactic category Expr) may contain embedded underscore symbols _ at places where identifiers are legal. Such an expression represents an anonymous function where subsequent occurrences of underscores denote successive parameters.
Note that one underscore references one input parameter, two underscores are for two different input parameters and so on.
With that said, you cannot use the placeholder twice and expect that they'll reference the same input parameter. That's not how it works in Scala and hence the compiler error.
// Not working
val rdd3 = rdd1.map((_, _.length))
The above is equivalent to the following:
// Not working
val rdd3 = rdd1.map { (a: String, b: String) => (a, b.length)) }
which is clearly incorrect as map expects a function of one input parameter.
Related
The scala documentation has a code example that includes the following line:
val numberFunc = numbers.foldLeft(List[Int]())_
What does the underscore after the method call mean?
It's a partially applied function. You only provide the first parameter to foldLeft (the initial value), but you don't provide the second one; you postpone it for later. In the docs you linked they do it in the next line, where they define squares:
val numberFunc = numbers.foldLeft(List[Int]())_
val squares = numberFunc((xs, x) => xs:+ x*x)
See that (xs, x) => xs:+ x*x, that's the missing second parameter which you omitted while defining numberFunc. If you had provided it right away, then numberFunc would not be a function - it would be the computed value.
So basically the whole thing can also be written as a one-liner in the curried form:
val squares = numbers.foldLeft(List[Int]())((xs, x) => xs:+ x*x)
However, if you want to be able to reuse foldLeft over and over again, having the same collection and initial value, but providing a different function every time, then it's very convinient to define a separate numbersFunc (as they did in the docs) and reuse it with different functions, e.g.:
val squares = numberFunc((xs, x) => xs:+ x*x)
val cubes = numberFunc((xs, x) => xs:+ x*x*x)
...
Note that the compiler error message is pretty straightforward in case you forget the underscore:
Error: missing argument list for method foldLeft in trait
LinearSeqOptimized Unapplied methods are only converted to functions
when a function type is expected. You can make this conversion
explicit by writing foldLeft _ or foldLeft(_)(_) instead of
foldLeft. val numberFunc = numbers.foldLeft(ListInt)
EDIT: Haha I just realized that they did the exact same thing with cubes in the documentation.
I don't know if it helps but I prefer this syntax
val numberFunc = numbers.foldLeft(List[Int]())(_)
then numberFunc is basically a delegate corresponding to an instance method (instance being numbers) waiting for a parameter. Which later comes to be a lambda expression in the scala documentation example
I have a function that takes a value and returns a list of pairs, pairUp.
and a key set, flightPass.keys
I want to write a for loop that runs pairUp for each value of flightPass.keys, and returns a big list of all these returned values.
val result:List[(Int, Int)] = pairUp(flightPass.keys.toSeq(0)).toList
for (flight<- flightPass.keys.toSeq.drop(1))
{val result:List[(Int, Int)] = result ++ pairUp(flight).toList}
I've tried a few different variations on this, always getting the error:
<console>:23: error: forward reference extends over definition of value result
for (flight<- flightPass.keys.toSeq.drop(1)) {val result:List[(Int, Int)] = result ++ pairUp(flight).toList}
^
I feel like this should work in other languages, so what am I doing wrong here?
First off, you've defined result as a val, which means it is immutable and can't be modified.
So if you want to apply "pairUp for each value of flightPass.keys", why not map()?
val result = flightPass.keys.map(pairUp) //add .toList if needed
A Scala method which converts a List of values into a List of Lists and then reduces them to a single List is called flatMap which is short for map then flatten. You would use it like this:
flightPass.keys.toSeq.flatMap(k => pairUp(k))
This will take each 'key' from flightPass.keys and pass it to pairUp (the mapping part), then take the resulting Lists from each call to pairUp and 'flatten' them, resulting in a single joined list.
I am new to Apache Spark and Scala. I am trying to understand something here: -
I have one array:
Companies= Array(
(Microsoft,478953),
(IBM,332042),
(JP Morgan,226003),
(Google,342033)
)
I wanted to divide this by another array, element by element:
Count = Array((Microsoft,4), (IBM,3), (JP Morgan,2), (Google,3))
I used this code :
val result: Array[(String, Double)] = wordMapCount
.zip(letterMapCount)
.map { case ((letter, wc), (_, lc)) => (letter, lc.toDouble / wc) }
From here: Divide Arrays
This works. However, I do not understand it. Why does zip require the second array and not the first one also the case matching how is that working here?
Why does zip require the second array and not the first one?
Because that's how zip works. It takes two separate RDD instances and maps one over the other to create pair of the first and second element:
def zip[U](other: RDD[U])(implicit arg0: ClassTag[U]): RDD[(T, U)]
case matching how is that working here
You have two tuples:
(Microsoft, 478953), (Microsoft,4)
What this partial function does decomposition of the tuple type via a call to Tuple2.unapply. This:
case ((letter, wc), (_, lc))
Means "extract the first argument (_1) from the first tuple into a fresh value named letter, and the second argument (_2) to a fresh value named wc. Same goes for the second tuple. And then, it creates a new tuple with letter as the first value and the division of lc and wc as the second argument.
I was trying to find line with maximum words, and i wrote the following lines, to run on spark-shell:
import java.lang.Math
val counts = textFile.map(line => line.split(" ").size).reduce((a, b) => Math.max(a, b))
But since, map is one to one , and flatMap is one to either zero or anything. So i tried replacing map with flatMap, in above code. But its giving error as:
<console>:24: error: type mismatch;
found : Int
required: TraversableOnce[?]
val counts = F1.flatMap(s => s.split(" ").size).reduce((a,b)=> Math.max(a,b))
If anybody could make me understand the reason, it will really be helpful.
flatMap must return an Iterable which is clearly not what you want. You do want a map because you want to map a line to the number of words, so you want a one-to-one function that takes a line and maps it to the number of words (though you could create a collection with one element, being the size of course...).
FlatMap is meant to associate a collection to an input, for instance if you wanted to map a line to all its words you would do:
val words = textFile.flatMap(x => x.split(" "))
and that would return an RDD[String] containing all the words.
In the end, map transforms an RDD of size N into another RDD of size N (e.g. your lines to their length) whereas flatMap transforms an RDD of size N into an RDD of size P (actually an RDD of size N into an RDD of size N made of collections, all these collections are then flattened to produce the RDD of size P).
P.S.: one last word that has nothing to do with your problem, it is more efficient to do (for a string s)
val nbWords = s.split(" ").length
than call .size(). Indeed, the split method returns an array of String and arrays do not have a size method. So when you call .size() you have an implicit conversion from Array[String] to SeqLike[String] which creates new objects. But Array[T] do have a length field so there's no conversion calling length. (It's a detail but I think it's good habit though).
Any use of map can be replaced by flatMap, but the function argument has to be changed to return a single-element List: textFile.flatMap(line => List(line.split(" ").size)). This isn't a good idea: it just makes your code less understandable and less efficient.
After reading Tired of Null Pointer Exceptions? Consider Using Java SE 8's Optional!'s part about why use flatMap() rather than Map(), I have realized the truly reason why flatMap() can not replace map() is that map() is not a special case of flatMap().
It's true that flatMap() means one-to-many, but that's not the only thing flatMap() does. It can also strip outer Stream() if put it simply.
See the definations of map and flatMap:
Stream<R> map(Function<? super T, ? extends R> mapper)
Stream<R> flatMap(Function<? super T, ? extends Stream<? extends R>> mapper)
the only difference is the type of returned value in inner function. What map() returned is "Stream<'what inner function returned'>", while what flatMap() returned is just "what inner function returned".
So you can say that flatMap() can kick outer Stream() away, but map() can't. This is the most difference in my opinion, and also why map() is not just a special case of flatMap().
ps:
If you really want to make one-to-one with flatMap, then you should change it into one-to-List(one). That means you should add an outer Stream() manually which will be stripped by flatMap() later. After that you'll get the same effect as using map().(Certainly, it's clumsy. So don't do like that.)
Here are examples for Java8, but the same as Scala:
use map():
list.stream().map(line -> line.split(" ").length)
deprecated use flatMap():
list.stream().flatMap(line -> Arrays.asList(line.split(" ").length).stream())
I have a for-comprehension with a generator from a Set[MyType]
This MyType has a lazy val variable called factsPair which returns a pair of sets:
(Set[MyFact], Set[MyFact]).
I wish to loop through all of them and unify the facts into one flattened pair (Set[MyFact], Set[MyFact]) as follows, however I am getting No implicit view available ... and not enough arguments for flatten: implicit (asTraversable ... errors. (I am a bit new to Scala so still trying to get used to the errors).
lazy val allFacts =
(for {
mytype <- mytypeList
} yield mytype.factsPair).flatten
What do I need to specify to flatten for this to work?
Scala flatten works on same types. You have a Seq[(Set[MyFact], Set[MyFact])], which can't be flattened.
I would recommend learning the foldLeft function, because it's very general and quite easy to use as soon as you get the hang of it:
lazy val allFacts = myTypeList.foldLeft((Set[MyFact](), Set[MyFact]())) {
case (accumulator, next) =>
val pairs1 = accumulator._1 ++ next.factsPair._1
val pairs2 = accumulator._2 ++ next.factsPair._2
(pairs1, pairs2)
}
The first parameter takes the initial element it will append the other elements to. We start with an empty Tuple[Set[MyFact], Set[MyFact]] initialized like this: (Set[MyFact](), Set[MyFact]()).
Next we have to specify the function that takes the accumulator and appends the next element to it and returns with the new accumulator that has the next element in it. Because of all the tuples, it doesn't look nice, but works.
You won't be able to use flatten for this, because flatten on a collection returns a collection, and a tuple is not a collection.
You can, of course, just split, flatten, and join again:
val pairs = for {
mytype <- mytypeList
} yield mytype.factsPair
val (first, second) = pairs.unzip
val allFacts = (first.flatten, second.flatten)
A tuple isn't traverable, so you can't flatten over it. You need to return something that can be iterated over, like a List, for example:
List((1,2), (3,4)).flatten // bad
List(List(1,2), List(3,4)).flatten // good
I'd like to offer a more algebraic view. What you have here can be nicely solved using monoids. For each monoid there is a zero element and an operation to combine two elements into one.
In this case, sets for a monoid: the zero element is an empty set and the operation is a union. And if we have two monoids, their Cartesian product is also a monoid, where the operations are defined pairwise (see examples on Wikipedia).
Scalaz defines monoids for sets as well as tuples, so we don't need to do anything there. We'll just need a helper function that combines multiple monoid elements into one, which is implemented easily using folding:
def msum[A](ps: Iterable[A])(implicit m: Monoid[A]): A =
ps.foldLeft(m.zero)(m.append(_, _))
(perhaps there already is such a function in Scala, I didn't find it). Using msum we can easily define
def pairs(ps: Iterable[MyType]): (Set[MyFact], Set[MyFact]) =
msum(ps.map(_.factsPair))
using Scalaz's implicit monoids for tuples and sets.