I want to rewrite a recursive function using pattern matching instead of if-else statements, but I am getting (correct) warning messages that some parts of the code are unreachable. In fact, I am getting wrong logic evaluation.
The function I am trying to re-write is:
def pascal(c: Int, r: Int): Int =
if (c == 0)
1
else if (c == r)
1
else
pascal(c - 1, r - 1) + pascal(c, r - 1)
This function works as expected. I re-wrote it as follows using pattern matching but now the function is not working as expected:
def pascal2 (c : Int, r : Int) : Int = c match {
case 0 => 1
case r => 1
case _ => pascal2(c - 1, r - 1) + pascal2(c, r - 1)
}
Where am I going wrong?
Main:
println("Pascal's Triangle")
for (row <- 0 to 10) {
for (col <- 0 to row)
print(pascal(col, row) + " ")
println()
}
The following statement "shadows" the variable r:
case r =>
That is to say, the "r" in that case statement is not, in fact, the "r" that you have defined above. It is it's own "r" which is equivalently equal to "c" because you are telling Scala to assign any value to some variable named "r."
Hence, what you really want is:
def pascal2(c: Int, r: Int): Int = c match{
case 0 => 1
case _ if c == r => 1
case _ => pascal2(c-1, r-1) + pascal2(c, r-1)
}
This is not, however tail recursive.
I fully agree with #wheaties and advice you to follow his directions. For sake of completeness I want to point out a few alternatives.
Alternative 1
You could write your own unapply:
def pascal(c: Int, r: Int): Int = {
object MatchesBoundary {
def unapply(i: Int) = if (i==0 || i==r) Some(i) else None
}
c match {
case MatchesBoundary(_) => 1
case _ => pascal(c-1, r-1) + pascal(c, r-1)
}
}
I would not claim that it improves readability in this case a lot. Just want to show the possibility to combine semantically similar cases (with identical/similar case bodies), which may be useful in more complex examples.
Alternative 2
There is also a possible solution, which exploits the fact that Scala's syntax for pattern matching only treats lower case variables as variables for a match. The following example shows what I mean by that:
def pascal(c: Int, r: Int): Int = {
val BoundaryL = 0
val BoundaryR = r
c match {
case BoundaryL => 1
case BoundaryR => 1
case _ => pascal(c-1, r-1) + pascal(c, r-1)
}
}
Since BoundaryL and BoundaryR start with upper case letters they are not treated as variables, but are used directly as matching object. Therefore the above works (while changing them to boundaryL and boundaryR would not, which btw also gives compiler warnings). This means that you could get your example to work simply by replacing r by R. Since this is a rather ugly solution I mention it only for educational purposes.
Related
I was wondering if there is some general method to convert a "normal" recursion with foo(...) + foo(...) as the last call to a tail-recursion.
For example (scala):
def pascal(c: Int, r: Int): Int = {
if (c == 0 || c == r) 1
else pascal(c - 1, r - 1) + pascal(c, r - 1)
}
A general solution for functional languages to convert recursive function to a tail-call equivalent:
A simple way is to wrap the non tail-recursive function in the Trampoline monad.
def pascalM(c: Int, r: Int): Trampoline[Int] = {
if (c == 0 || c == r) Trampoline.done(1)
else for {
a <- Trampoline.suspend(pascal(c - 1, r - 1))
b <- Trampoline.suspend(pascal(c, r - 1))
} yield a + b
}
val pascal = pascalM(10, 5).run
So the pascal function is not a recursive function anymore. However, the Trampoline monad is a nested structure of the computation that need to be done. Finally, run is a tail-recursive function that walks through the tree-like structure, interpreting it, and finally at the base case returns the value.
A paper from Rúnar Bjanarson on the subject of Trampolines: Stackless Scala With Free Monads
In cases where there is a simple modification to the value of a recursive call, that operation can be moved to the front of the recursive function. The classic example of this is Tail recursion modulo cons, where a simple recursive function in this form:
def recur[A](...):List[A] = {
...
x :: recur(...)
}
which is not tail recursive, is transformed into
def recur[A]{...): List[A] = {
def consRecur(..., consA: A): List[A] = {
consA :: ...
...
consrecur(..., ...)
}
...
consrecur(...,...)
}
Alexlv's example is a variant of this.
This is such a well known situation that some compilers (I know of Prolog and Scheme examples but Scalac does not do this) can detect simple cases and perform this optimisation automatically.
Problems combining multiple calls to recursive functions have no such simple solution. TMRC optimisatin is useless, as you are simply moving the first recursive call to another non-tail position. The only way to reach a tail-recursive solution is remove all but one of the recursive calls; how to do this is entirely context dependent but requires finding an entirely different approach to solving the problem.
As it happens, in some ways your example is similar to the classic Fibonnaci sequence problem; in that case the naive but elegant doubly-recursive solution can be replaced by one which loops forward from the 0th number.
def fib (n: Long): Long = n match {
case 0 | 1 => n
case _ => fib( n - 2) + fib( n - 1 )
}
def fib (n: Long): Long = {
def loop(current: Long, next: => Long, iteration: Long): Long = {
if (n == iteration)
current
else
loop(next, current + next, iteration + 1)
}
loop(0, 1, 0)
}
For the Fibonnaci sequence, this is the most efficient approach (a streams based solution is just a different expression of this solution that can cache results for subsequent calls). Now,
you can also solve your problem by looping forward from c0/r0 (well, c0/r2) and calculating each row in sequence - the difference being that you need to cache the entire previous row. So while this has a similarity to fib, it differs dramatically in the specifics and is also significantly less efficient than your original, doubly-recursive solution.
Here's an approach for your pascal triangle example which can calculate pascal(30,60) efficiently:
def pascal(column: Long, row: Long):Long = {
type Point = (Long, Long)
type Points = List[Point]
type Triangle = Map[Point,Long]
def above(p: Point) = (p._1, p._2 - 1)
def aboveLeft(p: Point) = (p._1 - 1, p._2 - 1)
def find(ps: Points, t: Triangle): Long = ps match {
// Found the ultimate goal
case (p :: Nil) if t contains p => t(p)
// Found an intermediate point: pop the stack and carry on
case (p :: rest) if t contains p => find(rest, t)
// Hit a triangle edge, add it to the triangle
case ((c, r) :: _) if (c == 0) || (c == r) => find(ps, t + ((c,r) -> 1))
// Triangle contains (c - 1, r - 1)...
case (p :: _) if t contains aboveLeft(p) => if (t contains above(p))
// And it contains (c, r - 1)! Add to the triangle
find(ps, t + (p -> (t(aboveLeft(p)) + t(above(p)))))
else
// Does not contain(c, r -1). So find that
find(above(p) :: ps, t)
// If we get here, we don't have (c - 1, r - 1). Find that.
case (p :: _) => find(aboveLeft(p) :: ps, t)
}
require(column >= 0 && row >= 0 && column <= row)
(column, row) match {
case (c, r) if (c == 0) || (c == r) => 1
case p => find(List(p), Map())
}
}
It's efficient, but I think it shows how ugly complex recursive solutions can become as you deform them to become tail recursive. At this point, it may be worth moving to a different model entirely. Continuations or monadic gymnastics might be better.
You want a generic way to transform your function. There isn't one. There are helpful approaches, that's all.
I don't know how theoretical this question is, but a recursive implementation won't be efficient even with tail-recursion. Try computing pascal(30, 60), for example. I don't think you'll get a stack overflow, but be prepared to take a long coffee break.
Instead, consider using a Stream or memoization:
val pascal: Stream[Stream[Long]] =
(Stream(1L)
#:: (Stream from 1 map { i =>
// compute row i
(1L
#:: (pascal(i-1) // take the previous row
sliding 2 // and add adjacent values pairwise
collect { case Stream(a,b) => a + b }).toStream
++ Stream(1L))
}))
The accumulator approach
def pascal(c: Int, r: Int): Int = {
def pascalAcc(acc:Int, leftover: List[(Int, Int)]):Int = {
if (leftover.isEmpty) acc
else {
val (c1, r1) = leftover.head
// Edge.
if (c1 == 0 || c1 == r1) pascalAcc(acc + 1, leftover.tail)
// Safe checks.
else if (c1 < 0 || r1 < 0 || c1 > r1) pascalAcc(acc, leftover.tail)
// Add 2 other points to accumulator.
else pascalAcc(acc, (c1 , r1 - 1) :: ((c1 - 1, r1 - 1) :: leftover.tail ))
}
}
pascalAcc(0, List ((c,r) ))
}
It does not overflow the stack but as on big row and column but Aaron mentioned it's not fast.
Yes it's possible. Usually it's done with accumulator pattern through some internally defined function, which has one additional argument with so called accumulator logic, example with counting length of a list.
For example normal recursive version would look like this:
def length[A](xs: List[A]): Int = if (xs.isEmpty) 0 else 1 + length(xs.tail)
that's not a tail recursive version, in order to eliminate last addition operation we have to accumulate values while somehow, for example with accumulator pattern:
def length[A](xs: List[A]) = {
def inner(ys: List[A], acc: Int): Int = {
if (ys.isEmpty) acc else inner(ys.tail, acc + 1)
}
inner(xs, 0)
}
a bit longer to code, but i think the idea i clear. Of cause you can do it without inner function, but in such case you should provide acc initial value manually.
I'm pretty sure it's not possible in the simple way you're looking for the general case, but it would depend on how elaborate you permit the changes to be.
A tail-recursive function must be re-writable as a while-loop, but try implementing for example a Fractal Tree using while-loops. It's possble, but you need to use an array or collection to store the state for each point, which susbstitutes for the data otherwise stored in the call-stack.
It's also possible to use trampolining.
It is indeed possible. The way I'd do this is to
begin with List(1) and keep recursing till you get to the
row you want.
Worth noticing that you can optimize it: if c==0 or c==r the value is one, and to calculate let's say column 3 of the 100th row you still only need to calculate the first three elements of the previous rows.
A working tail recursive solution would be this:
def pascal(c: Int, r: Int): Int = {
#tailrec
def pascalAcc(c: Int, r: Int, acc: List[Int]): List[Int] = {
if (r == 0) acc
else pascalAcc(c, r - 1,
// from let's say 1 3 3 1 builds 0 1 3 3 1 0 , takes only the
// subset that matters (if asking for col c, no cols after c are
// used) and uses sliding to build (0 1) (1 3) (3 3) etc.
(0 +: acc :+ 0).take(c + 2)
.sliding(2, 1).map { x => x.reduce(_ + _) }.toList)
}
if (c == 0 || c == r) 1
else pascalAcc(c, r, List(1))(c)
}
The annotation #tailrec actually makes the compiler check the function
is actually tail recursive.
It could be probably be further optimized since given that the rows are symmetric, if c > r/2, pascal(c,r) == pascal ( r-c,r).. but left to the reader ;)
I thought I could implement n+k patterns as an active pattern in scala via unapply, but it seems to fail with unspecified value parameter: k
object NPlusK {
def apply(n : Int, k : Int) = {
n + k
}
def unapply(n : Int, k : Int) = {
if (n > 0 && n > k) Some(n - k) else None
}
}
object Main {
def main(args: Array[String]): Unit = {
}
def fac(n: Int) : BigInt = {
n match {
case 0 => 1
case NPlusK(n, 1) => n * fac(n - 1)
}
}
}
Is it possible to implement n+k patterns in Scala and in that event how?
You should look at this question for a longer discussion, but here's a short adaptation for your specific case.
An unapply method can only take one argument, and must decide from that argument how to split it into two parts. Since there are multiple ways to divide some integer x into n and k such that x = n + k, you can't use an unapply for this.
You can get around it by creating a separate extractors for each k. Thus, instead of NplusK you'd have Nplus1, Nplus2, etc since there is exactly one way to get n from x such that x = n + 1.
case class NplusK(k: Int) {
def unapply(n: Int) = if (n > 0 && n > k) Some(n - k) else None
}
val Nplus1 = NplusK(1)
val Nplus1(n) = 5 // n = 4
So your match becomes:
n match {
case 0 => 1
case Nplus1(n) => n * fac(n - 1)
}
Deconstructor unapply does not work this way at all. It takes only one argument, the matched value, and returns an option on a tuple, with as many elements as there are arguments to the your pattern (NPlusK). That is, when you have
(n: Int) match {
...
case NPlusK(n, 1)
It will look for an unapply method with an Int (or supertype) argument. If there is such a method, and if the return type is a Tuple2 (as NPlusK appears with two arguments in the pattern), then it will try to match. Whatever subpattern there are inside NPlusK (here the variable n, and the constant 1), will not be passed to unapply in anyway (what do you expect if you write case NPlusK(NPlusK(1, x), NPlusK(1, y))?). Instead, if unapply returns some tuple, then each element of the tuple will be matched to the corresponding subpattern, here n which always matches, and 1 which will match if the value is equal to 1.
You could write
def unapply(n: Int) = if (n > 0) Some((n-1, 1)) else None.
That would match when your NPlusK(n, 1). But that would not match NPlusK(n, 2), nor NPlusK(1, n) (except if n is 2). This does not make much sense. A pattern should probably have only one possible match. NPlusK(x, y) can match n in many different ways.
What would work would be something Peano integers like, with Succ(n) matching n+1.
I've looked over a few implementations of Fibonacci function in Scala starting from a very simple one, to the more complicated ones.
I'm not entirely sure which one is the fastest. I'm leaning towards the impression that the ones that uses memoization is faster, however I wonder why Scala doesn't have a native memoization.
Can anyone enlighten me toward the best and fastest (and cleanest) way to write a fibonacci function?
The fastest versions are the ones that deviate from the usual addition scheme in some way. Very fast is the calculation somehow similar to a fast binary exponentiation based on these formulas:
F(2n-1) = F(n)² + F(n-1)²
F(2n) = (2F(n-1) + F(n))*F(n)
Here is some code using it:
def fib(n:Int):BigInt = {
def fibs(n:Int):(BigInt,BigInt) = if (n == 1) (1,0) else {
val (a,b) = fibs(n/2)
val p = (2*b+a)*a
val q = a*a + b*b
if(n % 2 == 0) (p,q) else (p+q,p)
}
fibs(n)._1
}
Even though this is not very optimized (e.g. the inner loop is not tail recursive), it will beat the usual additive implementations.
for me the simplest defines a recursive inner tail function:
def fib: Stream[Long] = {
def tail(h: Long, n: Long): Stream[Long] = h #:: tail(n, h + n)
tail(0, 1)
}
This doesn't need to build any Tuple objects for the zip and is easy to understand syntactically.
Scala does have memoization in the form of Streams.
val fib: Stream[BigInt] = 0 #:: 1 #:: fib.zip(fib.tail).map(p => p._1 + p._2)
scala> fib take 100 mkString " "
res22: String = 0 1 1 2 3 5 8 13 21 34 55 89 144 233 377 610 987 1597 2584 4181 ...
Stream is a LinearSeq so you might like to convert it to an IndexedSeq if you're doing a lot of fib(42) type calls.
However I would question what your use-case is for a fibbonaci function. It will overflow Long in less than 100 terms so larger terms aren't much use for anything. The smaller terms you can just stick in a table and look them up if speed is paramount. So the details of the computation probably don't matter much since for the smaller terms they're all quick.
If you really want to know the results for very big terms, then it depends on whether you just want one-off values (use Landei's solution) or, if you're making a sufficient number of calls, you may want to pre-compute the whole lot. The problem here is that, for example, the 100,000th element is over 20,000 digits long. So we're talking gigabytes of BigInt values which will crash your JVM if you try to hold them in memory. You could sacrifice accuracy and make things more manageable. You could have a partial-memoization strategy (say, memoize every 100th term) which makes a suitable memory / speed trade-off. There is no clear anwser for what is the fastest: it depends on your usage and resources.
This could work. it takes O(1) space O(n) time to calculate a number, but has no caching.
object Fibonacci {
def fibonacci(i : Int) : Int = {
def h(last : Int, cur: Int, num : Int) : Int = {
if ( num == 0) cur
else h(cur, last + cur, num - 1)
}
if (i < 0) - 1
else if (i == 0 || i == 1) 1
else h(1,2,i - 2)
}
def main(args: Array[String]){
(0 to 10).foreach( (x : Int) => print(fibonacci(x) + " "))
}
}
The answers using Stream (including the accepted answer) are very short and idiomatic, but they aren't the fastest. Streams memoize their values (which isn't necessary in iterative solutions), and even if you don't keep the reference to the stream, a lot of memory may be allocated and then immediately garbage-collected. A good alternative is to use an Iterator: it doesn't cause memory allocations, is functional in style, short and readable.
def fib(n: Int) = Iterator.iterate(BigInt(0), BigInt(1)) { case (a, b) => (b, a+b) }.
map(_._1).drop(n).next
A little simpler tail Recursive solution that can calculate Fibonacci for large values of n. The Int version is faster but is limited, when n > 46 integer overflow occurs
def tailRecursiveBig(n :Int) : BigInt = {
#tailrec
def aux(n : Int, next :BigInt, acc :BigInt) :BigInt ={
if(n == 0) acc
else aux(n-1, acc + next,next)
}
aux(n,1,0)
}
This has already been answered, but hopefully you will find my experience helpful. I had a lot of trouble getting my mind around scala infinite streams. Then, I watched Paul Agron's presentation where he gave very good suggestions: (1) implement your solution with basic Lists first, then if you are going to generify your solution with parameterized types, create a solution with simple types like Int's first.
using that approach I came up with a real simple (and for me, easy to understand solution):
def fib(h: Int, n: Int) : Stream[Int] = { h #:: fib(n, h + n) }
var x = fib(0,1)
println (s"results: ${(x take 10).toList}")
To get to the above solution I first created, as per Paul's advice, the "for-dummy's" version, based on simple lists:
def fib(h: Int, n: Int) : List[Int] = {
if (h > 100) {
Nil
} else {
h :: fib(n, h + n)
}
}
Notice that I short circuited the list version, because if i didn't it would run forever.. But.. who cares? ;^) since it is just an exploratory bit of code.
The code below is both fast and able to compute with high input indices. On my computer it returns the 10^6:th Fibonacci number in less than two seconds. The algorithm is in a functional style but does not use lists or streams. Rather, it is based on the equality \phi^n = F_{n-1} + F_n*\phi, for \phi the golden ratio. (This is a version of "Binet's formula".) The problem with using this equality is that \phi is irrational (involving the square root of five) so it will diverge due to finite-precision arithmetics if interpreted naively using Float-numbers. However, since \phi^2 = 1 + \phi it is easy to implement exact computations with numbers of the form a + b\phi for a and b integers, and this is what the algorithm below does. (The "power" function has a bit of optimization in it but is really just iteration of the "mult"-multiplication on such numbers.)
type Zphi = (BigInt, BigInt)
val phi = (0, 1): Zphi
val mult: (Zphi, Zphi) => Zphi = {
(z, w) => (z._1*w._1 + z._2*w._2, z._1*w._2 + z._2*w._1 + z._2*w._2)
}
val power: (Zphi, Int) => Zphi = {
case (base, ex) if (ex >= 0) => _power((1, 0), base, ex)
case _ => sys.error("no negative power plz")
}
val _power: (Zphi, Zphi, Int) => Zphi = {
case (t, b, e) if (e == 0) => t
case (t, b, e) if ((e & 1) == 1) => _power(mult(t, b), mult(b, b), e >> 1)
case (t, b, e) => _power(t, mult(b, b), e >> 1)
}
val fib: Int => BigInt = {
case n if (n < 0) => 0
case n => power(phi, n)._2
}
EDIT: An implementation which is more efficient and in a sense also more idiomatic is based on Typelevel's Spire library for numeric computations and abstract algebra. One can then paraphrase the above code in a way much closer to the mathematical argument (We do not need the whole ring-structure but I think it's "morally correct" to include it). Try running the following code:
import spire.implicits._
import spire.algebra._
case class S(fst: BigInt, snd: BigInt) {
override def toString = s"$fst + $snd"++"φ"
}
object S {
implicit object SRing extends Ring[S] {
def zero = S(0, 0): S
def one = S(1, 0): S
def plus(z: S, w: S) = S(z.fst + w.fst, z.snd + w.snd): S
def negate(z: S) = S(-z.fst, -z.snd): S
def times(z: S, w: S) = S(z.fst * w.fst + z.snd * w.snd
, z.fst * w.snd + z.snd * w.fst + z.snd * w.snd)
}
}
object Fibo {
val phi = S(0, 1)
val fib: Int => BigInt = n => (phi pow n).snd
def main(arg: Array[String]) {
println( fib(1000000) )
}
}
I know that you can do matching on lists in a way like
val list = List(1,2,3)
list match {
case head::tail => head
case _ => //whatever
}
so I started to wonder how this works. If I understand correctly, :: is just an operator, so what's to stop me from doing something like
4 match {
case x + 2 => x //I would expect x=2 here
}
If there is a way to create this kind of functionality, how is it done; if not, then why?
Pattern matching takes the input and decomposes it with an unapply function. So in your case, unapply(4) would have to return the two numbers that sum to 4. However, there are many pairs that sum to 4, so the function wouldn't know what to do.
What you need is for the 2 to be accessible to the unapply function somehow. A special case class that stores the 2 would work for this:
case class Sum(addto: Int) {
def unapply(i: Int) = Some(i - addto)
}
val Sum2 = Sum(2)
val Sum2(x) = 5 // x = 3
(It would be nice to be able to do something like val Sum(2)(y) = 5 for compactness, but Scala doesn't allow parameterized extractors; see here.)
[EDIT: This is a little silly, but you could actually do the following too:
val `2 +` = Sum(2)
val `2 +`(y) = 5 // y = 3
]
EDIT: The reason the head::tail thing works is that there is exactly one way to split the head from the tail of a list.
There's nothing inherently special about :: versus +: you could use + if you had a predetermined idea of how you wanted it to break a number. For example, if you wanted + to mean "split in half", then you could do something like:
object + {
def unapply(i: Int) = Some(i-i/2, i/2)
}
and use it like:
scala> val a + b = 4
a: Int = 2
b: Int = 2
scala> val c + d = 5
c: Int = 3
d: Int = 2
EDIT: Finally, this explains that, when pattern matching, A op B means the same thing as op(A,B), which makes the syntax look nice.
Matching with case head :: tail uses an infix operation pattern of the form p1 op p2 which gets translated to op(p1, p2) before doing the actual matching. (See API for ::)
The problem with + is the following:
While it is easy to add an
object + {
def unapply(value: Int): Option[(Int, Int)] = // ...
}
object which would do the matching, you may only supply one result per value. E.g.
object + {
def unapply(value: Int): Option[(Int, Int)] = value match {
case 0 => Some(0, 0)
case 4 => Some(3, 1)
case _ => None
}
Now this works:
0 match { case x + 0 => x } // returns 0
also this
4 match { case x + 1 => x } // returns 3
But this won’t and you cannot change it:
4 match { case x + 2 => x } // does not match
No problem for ::, though, because it is always defined what is head and what is tail of a list.
There are two ::s (pronounced "cons") in Scala. One is the operator on Lists and the other is a class, which represents a non empty list characterized by a head and a tail. So head :: tail is a constructor pattern, which has nothing to do with the operator.
I have a function:
def func(a: int, b: int, c: double): int
And I want to match various possible scenarios
Wherever c is 0, return b-a
Wherever c > 9, return 0
Wherever a=b return 0
And so on, before doing some more complex logic if none of the above are satisfied.
Do I have to match c separately first, or can I match on a,b,c, like _,_,0?
You can pattern match all described cases like this:
def func(a: Int, b: Int, c: Double) = (a, b, c) match {
case (a, b, 0) => b - a
case (a, b, c) if c > 9 || a == b => 0
case _ => 1 // add your logic here
}
Following on from my comments to Easy Angel's answer, I still feel this
if (c == 0)
b -a
else if (c > 9)
0
else if (a == b)
0
else
1 // your logic here
is clearer. Basically because there isn't really any pattern to match here.