Monads and Monoids: from daily Java to Big Data analytics in Scala

[Presented at JavaDay 2015]

Finally, after two decades of evolution, Java 8 made a step towards functional programming. What can Java learn from other mature functional languages? How to leverage obscure mathematical abstractions such as Monad or Monoid in practice? Usually people find it scary and difficult to understand. Oleksiy will explain these concepts in simple words to give a feeling of powerful tool applicable in many domains, from daily Java and Scala routines to Big Data analytics with Storm or Hadoop.

Transcript

1. Oleksiy Dyagilev
   

   View Slide

2. • lead software engineer in epam
   • working on scalable computing and data grids (GigaSpaces, Storm, Spark)
   • blog http://dyagilev.org
   

   View Slide

3. • Abstract Algebra (1900s?) and Category Theory (1940s)
   • Mathematicians study abstract structures and relationships between them
   

   View Slide

4. • Abstract Algebra (1900s?) and Category Theory (1940s)
   • Mathematicians study abstract structures and relationships between them
   • Early of 1990s, Eugenio Moggi described the general use of monad to
   structure programs
   

   View Slide

5. • Abstract Algebra (1900s?) and Category Theory (1940s)
   • Mathematicians study abstract structures and relationships between them
   • Early of 1990s, Eugenio Moggi described the general use of monad to
   structure programs
   • Early of 1990s, monad appeared in Haskell, a purely functional language.
   As well as other concepts such as Functor, Monoid, Arrow, etc
   

   View Slide

6. • Abstract Algebra (1900s?) and Category Theory (1940s)
   • Mathematicians study abstract structures and relationships between them
   • Early of 1990s, Eugenio Moggi described the general use of monad to
   structure programs
   • Early of 1990s, monad appeared in Haskell, a purely functional language.
   As well as other concepts such as Functor, Monoid, Arrow, etc
   • 2003, Martin Odersky creates Scala, a languages that unifies object-
   oriented and functional paradigms. Influenced by Haskell, Java, Erlang, etc.
   

   View Slide

7. • Abstract Algebra (1900s?) and Category Theory (1940s)
   • Mathematicians study abstract structures and relationships between them
   • Early of 1990s, Eugenio Moggi described the general use of monad to
   structure programs
   • Early of 1990s, monad appeared in Haskell, a purely functional language.
   As well as other concepts such as Functor, Monoid, Arrow, etc
   • 2003, Martin Odersky creates Scala, a languages that unifies object-
   oriented and functional paradigms. Influenced by Haskell, Java, Erlang, etc.
   • 2014, Java 8 released. Functional programming support – lambda, streams
   

   View Slide

8. • How abstractions from Math (Category Theory, Abstract Algebra) help in functional programming & Big Data
   • How to leverage them and become a better programmer
   

   View Slide

9. View Slide

10. View Slide

11. View Slide

12. View Slide

13. View Slide

14. View Slide

15. View Slide

16. View Slide

17. User user = findUser(userId);
    if (user != null) {
    Address address = user.getAddress();
    if (address != null) {
    String zipCode = address.getZipCode();
    if (zipCode != null) {
    City city = findCityByZipCode(zipCode);
    if (city != null) {
    return city.getName();
    }
    }
    }
    }
    return null;
    Example #1
    

    View Slide

18. Optional cityName = findUser(userId)
    .flatMap(user -> user.getAddress())
    .flatMap(address -> address.getZipCode())
    .flatMap(zipCode -> findCityByZipCode(zipCode))
    .map(city -> city.getName());
    which
    may not return a result.
    Refactored with Optional
    

    View Slide

19. Stream employees = companies.stream()
    .flatMap(company -> company.departments())
    .flatMap(department -> department.employees());
    Example #2
    which can return several values.
    

    View Slide

20. • container with a type M (e.g. Optional)
    • method M flatMap(T -> M) (e.g. flatMap(T -> Optional))
    • constructor to put T into M; same as a static method M unit(T) (e.g. Optional.of(x))
    

    View Slide

21. • container with a type M (e.g. Optional)
    • method M flatMap(T -> M) (e.g. flatMap(T -> Optional))
    • constructor to put T into M; same as a static method M unit(T) (e.g. Optional.of(x))
    M map(f) { return flatMap(x -> unit(f(x))) }
    Bonus: now we can define M map(T -> U)
    

    View Slide

22. • container with a type M (e.g. Optional)
    • method M flatMap(T -> M) (e.g. flatMap(T -> Optional))
    • constructor to put T into M; same as a static method M unit(T) (e.g. Optional.of(x))
    1. Left identity: unit(x).flatMap(f) = f(x)
    2. Right identity: m.flatMap(x -> unit(x)) = m
    3. Associativity: m.flatMap(f).flatMap(g) = m.flatMap(x -> f(x).flatMap(g)))
    M map(f) { return flatMap(x -> unit(f(x))) }
    Bonus: now we can define M map(T -> U)
    

    View Slide

23. Optional user = findUser(userId);
    Optional order = findOrder(orderId);
    Optional payment = findPayment(orderId);
    Optional placement = user
    .flatMap(u ->
    (order.flatMap(o ->
    (payment.map(p -> submitOrder(u, o, p))))));
    Java: looks ugly 
    

    View Slide

24. Optional user = findUser(userId);
    Optional order = findOrder(orderId);
    Optional payment = findPayment(orderId);
    Optional placement = user
    .flatMap(u ->
    (order.flatMap(o ->
    (payment.map(p -> submitOrder(u, o, p))))));
    Java: looks ugly 
    • Scala, for-comprehension
    • Haskell, do-notation
    • F#, computational expressions
    

    View Slide

25. Optional user = findUser(userId);
    Optional order = findOrder(orderId);
    Optional payment = findPayment(orderId);
    Optional placement = user
    .flatMap(u ->
    (order.flatMap(o ->
    (payment.map(p -> submitOrder(u, o, p))))));
    Java: looks ugly 
    val placement =
    for {
    u <- findUser(userId)
    o <- findOrder(orderId)
    p <- findPayment(orderId)
    } yield submitOrder(u, o, p)
    Scala: built-in monad Support 
    • Scala, for-comprehension
    • Haskell, do-notation
    • F#, computational expressions
    

    View Slide

26. View Slide

27. trait Parser[T] extends (String => ParseResult[T])
    sealed abstract class ParseResult[T]
    case class Success[T](result: T, rest: String) extends ParseResult[T]
    case class Failure() extends ParseResult[Nothing]
    val letter: Parser[Char] = …
    val digit: Parser[Char] = …
    val space: Parser[Char] = …
    def map[U](f: T => U): Parser[U] = parser { in => this(in) map f }
    def flatMap[U](f: T => Parser[U]): Parser[U] = parser { in => this(in) withNext f }
    def * : Parser[List[T]] = …
    

    View Slide

28. trait Parser[T] extends (String => ParseResult[T])
    sealed abstract class ParseResult[T]
    case class Success[T](result: T, rest: String) extends ParseResult[T]
    case class Failure() extends ParseResult[Nothing]
    val letter: Parser[Char] = …
    val digit: Parser[Char] = …
    val space: Parser[Char] = …
    def map[U](f: T => U): Parser[U] = parser { in => this(in) map f }
    def flatMap[U](f: T => Parser[U]): Parser[U] = parser { in => this(in) withNext f }
    def * : Parser[List[T]] = …
    val userParser = for {
    firstName <- letter.*
    _ <- space
    lastName <- letter.*
    _ <- space
    phone <- digit.*} yield User(firstName, lastName, phone)
    “John Doe 0671112222”
    

    View Slide

29. scala.Option java.Optional Absence of value
    scala.List java.Stream Multiple results
    scala.Future scalaz.Task java.CompletableFuture Asynchronous computations
    scalaz.Reader Read from shared environment
    scalaz.Writer Collect data in addition to computed values
    scalaz.State Maintain state
    scala.Try scalaz.\/ Handling failures
    

    View Slide

30. • Remove boilerplate
    • Modularity: separate computations from combination strategy
    • Composability: compose computations from simple ones
    • Improve maintainability
    • Better readability
    • Vocabulary
    

    View Slide

31. View Slide

32. New data
    All data Batch view
    Real-time view
    Data
    stream
    Batch processing
    Real-time processing
    Serving layer
    Query
    and merge
    

    View Slide

33. • Write job logic once and run on many Platforms(Hadoop, Storm)
    • Library authors talk about monoids all the time 
    

    View Slide

34. • Write job logic once and run on many Platforms(Hadoop, Storm)
    • Library authors talk about monoids all the time 
    def wordCount[P <: Platform[P]]
    (source: Producer[P, String], store: P#Store[String, Long]) =
    source.flatMap { sentence =>
    toWords(sentence).map(_ -> 1L)
    }.sumByKey(store)
    

    View Slide

35. • Write job logic once and run on many Platforms(Hadoop, Storm)
    • Library authors talk about monoids all the time 
    def wordCount[P <: Platform[P]]
    (source: Producer[P, String], store: P#Store[String, Long]) =
    source.flatMap { sentence =>
    toWords(sentence).map(_ -> 1L)
    }.sumByKey(store)
    def sumByKey(store: P#Store[K, V])(implicit semigroup: Semigroup[V]): Summer[P, K, V] = …
    

    View Slide

36. Given a set S and a binary operation +, we say that (, +) is a Semigroup if ∀ , , ∈ :
    • Closure: + ∈
    • Associativity: ( + ) + = + ( + )
    Monoid is a semigroup with identity element:
    • Identity: ∃ ∈ : + = + =
    • 3 * 2 (numbers under multiplication, 1 is the identity element)
    • 1 + 5 (numbers under addition, 0 is the identity element)
    • “ab” + “cd” (strings under concatenation, empty string is the identity element)
    • many more
    

    View Slide

37. Input
    data
    map
    map
    map
    map
    reduce
    reduce
    reduce
    output
    Having a sequence of elements of monoid M,
    we can reduce them into a final value
    Associativity ensure that we can parallelize computation(not exactly true)
    Identity allows to skip elements that don’t affect the result
    

    View Slide

38. Associativity: ( + ) + = + ( + )
    General Associativity Theorem
    https://proofwiki.org/wiki/General_Associativity_Theorem
    given:
    + + + + + + + ℎ
    you can place parentheses anywhere
    (( + ) + ( + )) + ( + + + ℎ )
    or
    ( + + + ) + ( + + + ℎ)
    

    View Slide

39. ℎ
    + + + +
    +
    +
    +
    

    View Slide

40. ℎ
    + + + +
    +
    +
    +
    

    View Slide

41. a b c d e f g h
    a + b + c + d + e + f
    Batch processing
    Real-time processing
    0
    1
    2
    3
    4
    5
    6
    7
    time
    1h now
    Real-time sums from 0,
    each batch
    Batch proc. recomputes
    total sum
    

    View Slide

42. a b c d e f g h
    a + b + c + d + e + f
    Batch processing
    Real-time processing
    0
    1
    2
    3
    4
    5
    6
    7
    time
    1h now
    Query
    and sum
    real-time + batch
    ( + + + + + ) + + ℎ
    (this is where Semigroup required)
    

    View Slide

43. View Slide

44. Bloom filter is a space-efficient probabilistic data structure to test presence of an element in a set
    0 0 0 0 0 0 0 0 0 0 0 0
    
    Operations:
    • Insert element
    • Query if element is present. The answer is either No or Maybe (false positives are possible)
    Consists of:
    • hash functions: ℎ1
    , ℎ2
    , … ℎ
    • bit array of bits
    

    View Slide

45. 0 0 1 0 0 0 0 1 0 1 0 0
    ℎ1
    () ℎ2
    () … ℎ
    ()
    
    set bit value to 1
    

    View Slide

46. 0 0 1 0 1 0 1 1 0 0 0 0
    ℎ1
    () ℎ2
    () … ℎ
    ()
    
    check if all bits are set to 1
    

    View Slide

47. 0 0 1 0 1 0 0 1 0 0 0 0
    Filter A: {1
    , 2
    , 3
    }
    1 0 1 0 0 0 0 0 1 0 0 0
    Filter B: {4
    , 5
    , 6
    }
    + OR
    1 0 1 0 1 0 0 1 1 0 0 0
    Filter A + B: {1
    , 2
    , 3
    , 4
    , 5
    , 6
    }
    

    View Slide

48. A few can be found in in Algebird (Abstract Algebra for Scala) https://github.com/twitter/algebird/
    • Bloom Filter
    • HyperLogLog
    • CountMinSketch
    • TopK
    • etc
    

    View Slide

49. • Monad is just a useful pattern in functional programming
    • You don’t need to understand Category Theory to use Monads
    • Once you grasp the idea, you will see this pattern everywhere
    • Semigroup (commutative) and monoid define properties useful in distributed computing and Lambda Architecture.
    • It’s all about associativity and commutativity. No nonsense!
    

    View Slide

50. View Slide