I've recently written the following combinations and permutations functions for an F# project, but I'm quite aware they're far from optimised.
/// Rotates a list by one place forward.
let rotate lst =
List.tail lst @ [List.head lst]
/// Gets all rotations of a list.
let getRotations lst =
let rec getAll lst i = if i = 0 then [] else lst :: (getAll (rotate lst) (i - 1))
getAll lst (List.length lst)
/// Gets all permutations (without repetition) of specified length from a list.
let rec getPerms n lst =
match n, lst with
| 0, _ -> seq [[]]
| _, [] -> seq []
| k, _ -> lst |> getRotations |> Seq.collect (fun r -> Seq.map ((@) [List.head r]) (getPerms (k - 1) (List.tail r)))
/// Gets all permutations (with repetition) of specified length from a list.
let rec getPermsWithRep n lst =
match n, lst with
| 0, _ -> seq [[]]
| _, [] -> seq []
| k, _ -> lst |> Seq.collect (fun x -> Seq.map ((@) [x]) (getPermsWithRep (k - 1) lst))
// equivalent: | k, _ -> lst |> getRotations |> Seq.collect (fun r -> List.map ((@) [List.head r]) (getPermsWithRep (k - 1) r))
/// Gets all combinations (without repetition) of specified length from a list.
let rec getCombs n lst =
match n, lst with
| 0, _ -> seq [[]]
| _, [] -> seq []
| k, (x :: xs) -> Seq.append (Seq.map ((@) [x]) (getCombs (k - 1) xs)) (getCombs k xs)
/// Gets all combinations (with repetition) of specified length from a list.
let rec getCombsWithRep n lst =
match n, lst with
| 0, _ -> seq [[]]
| _, [] -> seq []
| k, (x :: xs) -> Seq.append (Seq.map ((@) [x]) (getCombsWithRep (k - 1) lst)) (getCombsWithRep k xs)
Does anyone have any suggestions for how these functions (algorithms) can be sped up? I'm particularly interested in how the permutation (with and without repetition) ones can be improved. The business involving rotations of lists doesn't look too efficient to me in retrospect.
Here's my new implementation for the getPerms
function, inspired by Tomas's answer.
Unfortunately, it's not really any fast than the existing one. Suggestions?
let getPerms n lst =
let rec getPermsImpl acc n lst = seq {
match n, lst with
| k, x :: xs ->
if k > 0 then
for r in getRotations lst do
yield! getPermsImpl (List.head r :: acc) (k - 1) (List.tail r)
if k >= 0 then yield! getPermsImpl acc k []
| 0, [] -> yield acc
| _, [] -> ()
}
getPermsImpl List.empty n lst
If you want to write efficient functional code, then it is a good idea to avoid using the @
operator, because concatentation of lists is very inefficient.
Here is an example of how you can write a function to generate all combinations:
let rec combinations acc size set = seq {
match size, set with
| n, x::xs ->
if n > 0 then yield! combinations (x::acc) (n - 1) xs
if n >= 0 then yield! combinations acc n xs
| 0, [] -> yield acc
| _, [] -> () }
combinations [] 3 [1 .. 4]
The parameters of the function are:
acc
is used to remember elements that are already selected to be included in the combination (initially this is an empty list)size
is the remaining number of elements that we need to add to acc
(initially this is the required size of the combinations)set
is the set elements to choose fromThe function is implemented using a simple recursion. If we need to generate combinations of size n
then we can either add or don't add the current element, so we try to generate combinations using both options (first case) and add all of them to the generated sequence using yield!
. If we need 0 more elements, then we successfuly generated a combination (second case) and if we end with some other number but don't have any remaining elements to use then we cannot return anything (last case).
Combinations with repetition would be similar - the difference is that you don't need to remove the elements from the list (by using just xs
in the recursive calls) so there are more options of what to do.
I noticed that your updated getPerms function contains duplicates. Here's my crack at a dupe-free version. Hopefully the comments speak for themselves. The hardest part was writing an efficient distrib
function, because the concatenation operator has to be used somewhere. Luckily it's only used on small sublists, so the performance remains reasonable. My getAllPerms code below generates all permutations of [1..9] in around a quarter of a second, all 10-element permutations in around 2.5 seconds.
Edit: funny, I didn't look at Tomas' code, but his combinations function and my picks function are nearly identical.
// All ordered picks {x_i1, x_i2, .. , x_ik} of k out of n elements {x_1,..,x_n}
// where i1 < i2 < .. < ik
let picks n L =
let rec aux nleft acc L = seq {
match nleft,L with
| 0,_ -> yield acc
| _,[] -> ()
| nleft,h::t -> yield! aux (nleft-1) (h::acc) t
yield! aux nleft acc t }
aux n [] L
// Distribute an element y over a list:
// {x1,..,xn} --> {y,x1,..,xn}, {x1,y,x2,..,xn}, .. , {x1,..,xn,y}
let distrib y L =
let rec aux pre post = seq {
match post with
| [] -> yield (L @ [y])
| h::t -> yield (pre @ y::post)
yield! aux (pre @ [h]) t }
aux [] L
// All permutations of a single list = the head of a list distributed
// over all permutations of its tail
let rec getAllPerms = function
| [] -> Seq.singleton []
| h::t -> getAllPerms t |> Seq.collect (distrib h)
// All k-element permutations out of n elements =
// all permutations of all ordered picks of length k combined
let getPerms2 n lst = picks n lst |> Seq.collect getAllPerms
Edit: more code in response to comments
// Generates the cartesian outer product of a list of sequences LL
let rec outerProduct = function
| [] -> Seq.singleton []
| L::Ls -> L |> Seq.collect (fun x ->
outerProduct Ls |> Seq.map (fun L -> x::L))
// Generates all n-element combination from a list L
let getPermsWithRep2 n L =
List.replicate n L |> outerProduct
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With