You can think of this function as choosing, for each place in between two list elements, whether to include a split there. So for starters, there should be 2n-1 partitions for an n-element list: you can use that as a quick sanity check on a possible solution.
One good way to model non-determinism is with the list monad (or equivalently with list comprehensions), so let's do it that way.
First, let's write the type and a base case:
separate :: [a] -> [[[a]]]
separate [] = [[]]
There is a single way to separate an empty list: the empty list itself, with no possibility of splits. Easy enough.
Now, given we have one element and a list of remaining elements, one thing we'll need for sure is a list of all the ways to split the remaining elements:
separate :: [a] -> [[[a]]]
separate [] = [[]]
separate (x:xs) = let recur = separate xs
in undefined -- TODO
Here's where the interesting stuff starts. As I said, you can view this as choosing, for each item, whether to put a split after it. Two choices means concatenating together two lists, so let's do that:
separate :: [a] -> [[[a]]]
separate [] = [[]]
separate (x:xs) = let recur = separate xs
split = undefined -- TODO
noSplit = undefined -- TODO
in split ++ noSplit
Now, how do we introduce a split after the item x
? We do it by, for each partition in recur
, adding [x]
to the front of it as a new partition:
separate :: [a] -> [[[a]]]
separate [] = [[]]
separate (x:xs) = let recur = separate xs
split = do
partition <- recur
return $ [x] : partition
noSplit = undefined -- TODO
in split ++ noSplit
What about not splitting? Pretty similar! For each partition in recur
, we add x
to the front of the first sub-partition:
separate :: [a] -> [[[a]]]
separate [] = [[]]
separate (x:xs) = let recur = separate xs
split = do
partition <- recur
return $ [x] : partition
noSplit = do
(y:ys) <- recur
return $ (x:y):ys
in split ++ noSplit
And with that, we're done:
*Temp> separate "123"
[["1","2","3"],["1","23"],["12","3"],["123"]]