If I call this function with a very high initial currentReflection value I get a stack overflow exception, which indicates that the function is not tail-recursive (correct?). My understanding was that as long as the recursive call was the final computation of the function then it should be compiler-optimized as a tail-recursive function to reuse the current stack frame. Anyone know why this isn't the case here?
let rec traceColorAt intersection ray currentReflection =
// some useful values to compute at the start
let matrix = intersection.sphere.transformation |> transpose |> invert
let transNormal = matrix.Transform(intersection.normal) |> norm
let hitPoint = intersection.point
let ambient = ambientColorAt intersection
let specular = specularColorAt intersection hitPoint transNormal
let diffuse = diffuseColorAt intersection hitPoint transNormal
let primaryColor = ambient + diffuse + specular
if currentReflection = 0 then
primaryColor
else
let reflectDir = (ray.direction - 2.0 * norm ((Vector3D.DotProduct(ray.direction, intersection.normal)) * intersection.normal))
let newRay = { origin=intersection.point; direction=reflectDir }
let intersections = castRay newRay scene
match intersections with
| [] -> primaryColor
| _ ->
let newIntersection = List.minBy(fun x -> x.t) intersections
let reflectivity = intersection.sphere.material.reflectivity
primaryColor + traceColorAt newIntersection newRay (currentReflection - 1) * reflectivity
The recursive call to traceColorAt appears as part of a larger expression. This prevents tail call optimization because further computation is necessary after traceColorAt returns.
To convert this function to be tail recursive, you could add an additional accumulator parameter for primaryColor. The outermost call to traceColorAt would pass the "zero" value for primaryColor (black?) and each recursive call would sum in the adjustment it computes, e.g. the code would look something like:
let rec traceColorAt intersection ray currentReflection primaryColor
...
let newPrimaryColor = primaryColor + ambient + diffuse + specular
...
match intersections with
| [] -> newPrimaryColor
| _ ->
...
traceColorAt newIntersection newRay ((currentReflection - 1) * reflectivity) newPrimaryColor
If you wish to hide the extra parameter from callers, introduce a helper function that performs the bulk of the work and call that from traceColorAt.
Tail recursion works if the function would simply return the result of another function. In this case, you have primaryColor + traceColorAt(...), which means that it is not simply returning the value of the function-- it's also adding something to it.
You could fix this by passing the current accumulated color as a parameter.
Related
I'm a beginner in functional programming but I'm famaliar with imperative programming. I'm having trouble translating a piece of cpp code involving updatating two objects at the same time (context is n-body simulation).
It's roughly like this in c++:
for (Particle &i: particles) {
for (Particle &j: particles) {
collide(i, j) // function that mutates particles i and j
}
}
I'm translating this to Ocaml, with immutable objects and immutable Lists. The difficult part is that I need to replace two objects at the same time. So far I have this:
List.map (fun i ->
List.map (fun j ->
let (new_i, new_j) = collide(i, j) in // function that returns new particles i, j
// how do i update particles with new i, j?
) particles
) particles
How do I replace both objects in the List at the same time?
The functional equivalent of the imperative code is just as simple as,
let nbody f xs =
List.map (fun x -> List.fold_left f x xs) xs
It is a bit more generic, as a I abstracted the collide function and made it a parameter. The function f takes two bodies and returns the state of the first body as affected by the second body. For example, we can implement the following symbolic collide function,
let symbolic x y = "f(" ^ x ^ "," ^ y ^ ")"
so that we can see the result and associativity of the the collide function application,
# nbody symbolic [
"x"; "y"; "z"
];;
- : string list =
["f(f(f(x,x),y),z)"; "f(f(f(y,x),y),z)"; "f(f(f(z,x),y),z)"]
So, the first element of the output list is the result of collision of x with x itself, then with y, then with z. The second element is the result of collision of y with x, and y, and z. And so on.
Obviously the body shall not collide with itself, but this could be easily fixed by either modifying the collide function or by filtering the input list to List.fold and removing the currently being computed element. This is left as an exercise.
List.map returns a new list. The function you supply to List.map may transform the elements from one type to another or just apply some operation on the same type.
For example, let's assume you start with a list of integer tuples
let int_tuples = [(1, 3); (4, 3); (8, 2)];;
and let's assume that your update function takes an integer tuple and doubles the integers:
let update (i, j) = (i * 2, j * 2) (* update maybe your collide function *)
If you now do:
let new_int_tuples = List.map update int_tuples
You'll get
(* [(2, 6); (8, 6); (16, 4)] *)
Hope this helps
I'm learning functional programming with F#, and I want to write a function that will generate a sequence for me.
There is a some predetermined function for transforming a value, and in the function I need to write there should be two inputs - the starting value and the length of the sequence. Sequence starts with the initial value, and each following item is a result of applying the transforming function to the previous value in the sequence.
In C# I would normally write something like that:
public static IEnumerable<double> GenerateSequence(double startingValue, int n)
{
double TransformValue(double x) => x * 0.9 + 2;
yield return startingValue;
var returnValue = startingValue;
for (var i = 1; i < n; i++)
{
returnValue = TransformValue(returnValue);
yield return returnValue;
}
}
As I tried to translate this function to F#, I made this:
let GenerateSequence startingValue n =
let transformValue x =
x * 0.9 + 2.0
seq {
let rec repeatableFunction value n =
if n = 1 then
transformValue value
else
repeatableFunction (transformValue value) (n-1)
yield startingValue
for i in [1..n-1] do
yield repeatableFunction startingValue i
}
There are two obvious problems with this implementation.
First is that because I tried to avoid making a mutable value (analogy of returnValue variable in C# implementation), I didn't reuse values of former computations while generating sequence. This means that for the 100th element of the sequence I have to make additional 99 calls of the transformValue function instead of just one (as I did in C# implementation). This reeks with extremely bad performance.
Second is that the whole function does not seem to be written in accordance with Functional Programming. I am pretty sure that there are more elegant and compact implementation. I suspect that Seq.fold or List.fold or something like that should have been used here, but I'm still not able to grasp how to effectively use them.
So the question is: how to re-write the GenerateSequence function in F# so it would be in Functional Programming style and have a better performance?
Any other advice would also be welcomed.
The answer from #rmunn shows a rather nice solution using unfold. I think there are other two options worth considering, which are actually just using a mutable variable and using a recursive sequence expression. The choice is probably a matter of personal preference. The two other options look like this:
let generateSequenceMutable startingValue n = seq {
let transformValue x = x * 0.9 + 2.0
let mutable returnValue = startingValue
for i in 1 .. n do
yield returnValue
returnValue <- transformValue returnValue }
let generateSequenceRecursive startingValue n =
let transformValue x = x * 0.9 + 2.0
let rec loop value i = seq {
if i < n then
yield value
yield! loop (transformValue value) (i + 1) }
loop startingValue 0
I modified your logic slightly so that I do not have to yield twice - I just do one more step of the iteration and yield before updating the value. This makes the generateSequenceMutable function quite straightforward and easy to understand. The generateSequenceRecursive implements the same logic using recursion and is also fairly nice, but I find it a bit less clear.
If you wanted to use one of these versions and generate an infinite sequence from which you can then take as many elements as you need, you can just change for to while in the first case or remove the if in the second case:
let generateSequenceMutable startingValue n = seq {
let transformValue x = x * 0.9 + 2.0
let mutable returnValue = startingValue
while true do
yield returnValue
returnValue <- transformValue returnValue }
let generateSequenceRecursive startingValue n =
let transformValue x = x * 0.9 + 2.0
let rec loop value i = seq {
yield value
yield! loop (transformValue value) (i + 1) }
loop startingValue 0
If I was writing this, I'd probably go either with the mutable variable or with unfold. Mutation may be "generally evil" but in this case, it is a localized mutable variable that is not breaking referential transparency in any way, so I don't think it's harmful.
Your description of the problem was excellent: "Sequence starts with the initial value, and each following item is a result of applying the transforming function to the previous value in the sequence."
That is a perfect description of the Seq.unfold method. It takes two parameters: the initial state and a transformation function, and returns a sequence where each value is calculated from the previous state. There are a few subtleties involved in using Seq.unfold which the rather terse documentation may not explain very well:
Seq.unfold expects the transformation function, which I'll call f from now on, to return an option. It should return None if the sequence should end, or Some (...) if there's another value left in the sequence. You can create infinite sequences this way if you never return None; infinite sequences are perfectly fine since F# evaluates sequences lazily, but you do need to be careful not to ever loop over the entirely of an infinite sequence. :-)
Seq.unfold also expects that if f returns Some (...), it will return not just the next value, but a tuple of the next value and the next state. This is shown in the Fibonacci example in the documentation, where the state is actually a tuple of the current value and the previous value, which will be used to calculate the next value shown. The documentation example doesn't make that very clear, so here's what I think is a better example:
let infiniteFibonacci = (0,1) |> Seq.unfold (fun (a,b) ->
// a is the value produced *two* iterations ago, b is previous value
let c = a+b
Some (c, (b,c))
)
infiniteFibonacci |> Seq.take 5 |> List.ofSeq // Returns [1; 2; 3; 5; 8]
let fib = seq {
yield 0
yield 1
yield! infiniteFibonacci
}
fib |> Seq.take 7 |> List.ofSeq // Returns [0; 1; 1; 2; 3; 5; 8]
And to get back to your GenerateSequence question, I would write it like this:
let GenerateSequence startingValue n =
let transformValue x =
let result = x * 0.9 + 2.0
Some (result, result)
startingValue |> Seq.unfold transformValue |> Seq.take n
Or if you need to include the starting value in the sequence:
let GenerateSequence startingValue n =
let transformValue x =
let result = x * 0.9 + 2.0
Some (result, result)
let rest = startingValue |> Seq.unfold transformValue |> Seq.take n
Seq.append (Seq.singleton startingValue) rest
The difference between Seq.fold and Seq.unfold
The easiest way to remember whether you want to use Seq.fold or Seq.unfold is to ask yourself which of these two statements is true:
I have a list (or array, or sequence) of items, and I want to produce a single result value by running a calculation repeatedly on pairs of items in the list. For example, I want to take the product of this whole series of numbers. This is a fold operation: I take a long list and "compress" it (so to speak) until it's a single value.
I have a single starting value and a function to produce the next value from the current value, and I want to end up with a list (or sequence, or array) of values. This is an unfold operation: I take a small starting value and "expand" it (so to speak) until it's a whole list of values.
I am attempting to create a new time series (Series[DateTime,float]) from an existing one of the same type, where the map to the new Series is recursive - for example:
NewSeries_T = NewSeries_T-1 + constant * OldSeries_T;
I have "NewSeries_0 = 1", as an initialization value for the new series.
I'm attempting to write a Series.map function that will do the job - I've got as far as the following non-working code, but I can't figure out the recursive part:
let rec newSeries = existingSeries |> Series.map (fun k v ->
match k.Equals(initDate) with
| true -> 1
| false -> newSeries.LastValue() + constant * v
)
So, I think the trick is, how do I allow the function access to the "previous" value in the series to build this up recursively?
Edit - Moved to answer below.
Based on Fyodor's reccomendation - the Series.scanValues does exactly what I need:
let initalEntry = Series([initDate], [init])
let newSeries=
existingSeries
|> Series.filter (fun k v -> k.Equals(initDate) = false)
|> Series.scanValues (fun n x-> lambda * n + (1.0 - lambda) * x ) init
newSeries.Merge(initalEntry)
I took away the first value, as I want this to return the "init" value at the start of the series, and merged this back at the end.
I'm new to ocaml and tryin to write a continuation passing style function but quite confused what value i need to pass into additional argument on k
for example, I can write a recursive function that returns true if all elements of the list is even, otherwise false.
so its like
let rec even list = ....
on CPS, i know i need to add one argument to pass function
so like
let rec evenk list k = ....
but I have no clue how to deal with this k and how does this exactly work
for example for this even function, environment looks like
val evenk : int list -> (bool -> ’a) -> ’a = <fun>
evenk [4; 2; 12; 5; 6] (fun x -> x) (* output should give false *)
The continuation k is a function that takes the result from evenk and performs "the rest of the computation" and produces the "answer". What type the answer has and what you mean by "the rest of the computation" depends on what you are using CPS for. CPS is generally not an end in itself but is done with some purpose in mind. For example, in CPS form it is very easy to implement control operators or to optimize tail calls. Without knowing what you are trying to accomplish, it's hard to answer your question.
For what it is worth, if you are simply trying to convert from direct style to continuation-passing style, and all you care about is the value of the answer, passing the identity function as the continuation is about right.
A good next step would be to implement evenk using CPS. I'll do a simpler example.
If I have the direct-style function
let muladd x i n = x + i * n
and if I assume CPS primitives mulk and addk, I can write
let muladdk x i n k =
let k' product = addk x product k in
mulk i n k'
And you'll see that the mulptiplication is done first, then it "continues" with k', which does the add, and finally that continues with k, which returns to the caller. The key idea is that within the body of muladdk I allocated a fresh continuation k' which stands for an intermediate point in the multiply-add function. To make your evenk work you will have to allocate at least one such continuation.
I hope this helps.
Whenever I've played with CPS, the thing passed to the continuation is just the thing you would normally return to the caller. In this simple case, a nice "intuition lubricant" is to name the continuation "return".
let rec even list return =
if List.length list = 0
then return true
else if List.hd list mod 2 = 1
then return false
else even (List.tl list) return;;
let id = fun x -> x;;
Example usage: "even [2; 4; 6; 8] id;;".
Since you have the invocation of evenk correct (with the identity function - effectively converting the continuation-passing-style back to normal style), I assume that the difficulty is in defining evenk.
k is the continuation function representing the rest of the computation and producing a final value, as Norman said. So, what you need to do is compute the result of v of even and pass that result to k, returning k v rather than just v.
You want to give as input the result of your function as if it were not written with continuation passing style.
Here is your function which tests whether a list has only even integers:
(* val even_list : int list -> bool *)
let even_list input = List.for_all (fun x -> x mod 2=0) input
Now let's write it with a continuation cont:
(* val evenk : int list -> (bool -> 'a) -> 'a *)
let evenk input cont =
let result = even_list input in
(cont result)
You compute the result your function, and pass resultto the continuation ...
Given this algorithm, I would like to know if there exists an iterative version. Also, I want to know if the iterative version can be faster.
This some kind of pseudo-python...
the algorithm returns a reference to root of the tree
make_tree(array a)
if len(a) == 0
return None;
node = pick a random point from the array
calculate distances of the point against the others
calculate median of such distances
node.left = make_tree(subset of the array, such that the distance of points is lower to the median of distances)
node.right = make_tree(subset, such the distance is greater or equal to the median)
return node
A recursive function with only one recursive call can usually be turned into a tail-recursive function without too much effort, and then it's trivial to convert it into an iterative function. The canonical example here is factorial:
# naïve recursion
def fac(n):
if n <= 1:
return 1
else:
return n * fac(n - 1)
# tail-recursive with accumulator
def fac(n):
def fac_helper(m, k):
if m <= 1:
return k
else:
return fac_helper(m - 1, m * k)
return fac_helper(n, 1)
# iterative with accumulator
def fac(n):
k = 1
while n > 1:
n, k = n - 1, n * k
return k
However, your case here involves two recursive calls, and unless you significantly rework your algorithm, you need to keep a stack. Managing your own stack may be a little faster than using Python's function call stack, but the added speed and depth will probably not be worth the complexity. The canonical example here would be the Fibonacci sequence:
# naïve recursion
def fib(n):
if n <= 1:
return 1
else:
return fib(n - 1) + fib(n - 2)
# tail-recursive with accumulator and stack
def fib(n):
def fib_helper(m, k, stack):
if m <= 1:
if stack:
m = stack.pop()
return fib_helper(m, k + 1, stack)
else:
return k + 1
else:
stack.append(m - 2)
return fib_helper(m - 1, k, stack)
return fib_helper(n, 0, [])
# iterative with accumulator and stack
def fib(n):
k, stack = 0, []
while 1:
if n <= 1:
k = k + 1
if stack:
n = stack.pop()
else:
break
else:
stack.append(n - 2)
n = n - 1
return k
Now, your case is a lot tougher than this: a simple accumulator will have difficulties expressing a partly-built tree with a pointer to where a subtree needs to be generated. You'll want a zipper -- not easy to implement in a not-really-functional language like Python.
Making an iterative version is simply a matter of using your own stack instead of the normal language call stack. I doubt the iterative version would be faster, as the normal call stack is optimized for this purpose.
The data you're getting is random so the tree can be an arbitrary binary tree. For this case, you can use a threaded binary tree, which can be traversed and built w/o recursion and no stack. The nodes have a flag that indicate if the link is a link to another node or how to get to the "next node".
From http://en.wikipedia.org/wiki/Threaded_binary_tree
Depending on how you define "iterative", there is another solution not mentioned by the previous answers. If "iterative" just means "not subject to a stack overflow exception" (but "allowed to use 'let rec'"), then in a language that supports tail calls, you can write a version using continuations (rather than an "explicit stack"). The F# code below illustrates this. It is similar to your original problem, in that it builds a BST out of an array. If the array is shuffled randomly, the tree is relatively balanced and the recursive version does not create too deep a stack. But turn off shuffling, and the tree gets unbalanced, and the recursive version stack-overflows whereas the iterative-with-continuations version continues along happily.
#light
open System
let printResults = false
let MAX = 20000
let shuffleIt = true
// handy helper function
let rng = new Random(0)
let shuffle (arr : array<'a>) = // '
let n = arr.Length
for x in 1..n do
let i = n-x
let j = rng.Next(i+1)
let tmp = arr.[i]
arr.[i] <- arr.[j]
arr.[j] <- tmp
// Same random array
let sampleArray = Array.init MAX (fun x -> x)
if shuffleIt then
shuffle sampleArray
if printResults then
printfn "Sample array is %A" sampleArray
// Tree type
type Tree =
| Node of int * Tree * Tree
| Leaf
// MakeTree1 is recursive
let rec MakeTree1 (arr : array<int>) lo hi = // [lo,hi)
if lo = hi then
Leaf
else
let pivot = arr.[lo]
// partition
let mutable storeIndex = lo + 1
for i in lo + 1 .. hi - 1 do
if arr.[i] < pivot then
let tmp = arr.[i]
arr.[i] <- arr.[storeIndex]
arr.[storeIndex] <- tmp
storeIndex <- storeIndex + 1
Node(pivot, MakeTree1 arr (lo+1) storeIndex, MakeTree1 arr storeIndex hi)
// MakeTree2 has all tail calls (uses continuations rather than a stack, see
// http://lorgonblog.spaces.live.com/blog/cns!701679AD17B6D310!171.entry
// for more explanation)
let MakeTree2 (arr : array<int>) lo hi = // [lo,hi)
let rec MakeTree2Helper (arr : array<int>) lo hi k =
if lo = hi then
k Leaf
else
let pivot = arr.[lo]
// partition
let storeIndex = ref(lo + 1)
for i in lo + 1 .. hi - 1 do
if arr.[i] < pivot then
let tmp = arr.[i]
arr.[i] <- arr.[!storeIndex]
arr.[!storeIndex] <- tmp
storeIndex := !storeIndex + 1
MakeTree2Helper arr (lo+1) !storeIndex (fun lacc ->
MakeTree2Helper arr !storeIndex hi (fun racc ->
k (Node(pivot,lacc,racc))))
MakeTree2Helper arr lo hi (fun x -> x)
// MakeTree2 never stack overflows
printfn "calling MakeTree2..."
let tree2 = MakeTree2 sampleArray 0 MAX
if printResults then
printfn "MakeTree2 yields"
printfn "%A" tree2
// MakeTree1 might stack overflow
printfn "calling MakeTree1..."
let tree1 = MakeTree1 sampleArray 0 MAX
if printResults then
printfn "MakeTree1 yields"
printfn "%A" tree1
printfn "Trees are equal: %A" (tree1 = tree2)
Yes it is possible to make any recursive algorithm iterative. Implicitly, when you create a recursive algorithm each call places the prior call onto the stack. What you want to do is make the implicit call stack into an explicit one. The iterative version won't necessarily be faster, but you won't have to worry about a stack overflow. (do I get a badge for using the name of the site in my answer?
While it is true in the general sense that directly converting a recursive algorithm into an iterative one will require an explicit stack, there is a specific sub-set of algorithms which render directly in iterative form (without the need for a stack). These renderings may not have the same performance guarantees (iterating over a functional list vs recursive deconstruction), but they do often exist.
Here is stack based iterative solution (Java):
public static Tree builtBSTFromSortedArray(int[] inputArray){
Stack toBeDone=new Stack("sub trees to be created under these nodes");
//initialize start and end
int start=0;
int end=inputArray.length-1;
//keep memoy of the position (in the array) of the previously created node
int previous_end=end;
int previous_start=start;
//Create the result tree
Node root=new Node(inputArray[(start+end)/2]);
Tree result=new Tree(root);
while(root!=null){
System.out.println("Current root="+root.data);
//calculate last middle (last node position using the last start and last end)
int last_mid=(previous_start+previous_end)/2;
//*********** add left node to the previously created node ***********
//calculate new start and new end positions
//end is the previous index position minus 1
end=last_mid-1;
//start will not change for left nodes generation
start=previous_start;
//check if the index exists in the array and add the left node
if (end>=start){
root.left=new Node(inputArray[((start+end)/2)]);
System.out.println("\tCurrent root.left="+root.left.data);
}
else
root.left=null;
//save previous_end value (to be used in right node creation)
int previous_end_bck=previous_end;
//update previous end
previous_end=end;
//*********** add right node to the previously created node ***********
//get the initial value (inside the current iteration) of previous end
end=previous_end_bck;
//start is the previous index position plus one
start=last_mid+1;
//check if the index exists in the array and add the right node
if (start<=end){
root.right=new Node(inputArray[((start+end)/2)]);
System.out.println("\tCurrent root.right="+root.right.data);
//save the created node and its index position (start & end) in the array to toBeDone stack
toBeDone.push(root.right);
toBeDone.push(new Node(start));
toBeDone.push(new Node(end));
}
//*********** update the value of root ***********
if (root.left!=null){
root=root.left;
}
else{
if (toBeDone.top!=null) previous_end=toBeDone.pop().data;
if (toBeDone.top!=null) previous_start=toBeDone.pop().data;
root=toBeDone.pop();
}
}
return result;
}