# Create Binary, Balanced Tree

I wrote a function, which takes a list of Integers, and then makes a balanced binary tree.

Please critique it. I'd like to know know how to make it more concise and idiomatic.

-- Note that the Integer represents the height of the Tree
data Tree a = Leaf
| Node Integer (Tree a) a (Tree a)
deriving (Show)-- where Integer is the height (bottom = 0)

-- make balanced, binary tree (height diff is <= 1)
-- Note - I would've liked to have kept foldTree point-free,
--        but I'm not sure how to do that since I need xs for treeHeight
foldTree :: [a] -> Tree a
foldTree xs = (foldingFn . zip [0..]) xs
where foldingFn = foldr (\(i, elem) acc -> if (odd i) then insertFreeOrLeft  treeHeight elem acc
else            insertFreeOrRight treeHeight elem acc) Leaf
treeHeight = getBinTreeHt xs

-- get Binary Tree Height (used to start making the Tree)
getBinTreeHt :: [a] -> Integer
getBinTreeHt = floor . (logBase 2) . fromIntegral . length

-- insert where there's a Leaf, otherwise choose Left
insertFreeOrLeft :: Integer -> a -> Tree a -> Tree a
insertFreeOrLeft index x Leaf                    = Node index  Leaf x Leaf
insertFreeOrLeft _ x (Node level Leaf val right) = Node level (Node (level-1) Leaf x Leaf) val right
insertFreeOrLeft _ x (Node level left val Leaf)  = Node level left val (Node (level-1) Leaf x Leaf)
insertFreeOrLeft _ x (Node level left val right) = Node level (insertFreeOrLeft (level-1) x left) val right

-- insert where there's a Leaf, otherwise choose Right
insertFreeOrRight :: Integer -> a -> Tree a -> Tree a
insertFreeOrRight index x Leaf                    = Node index  Leaf x Leaf
insertFreeOrRight _ x (Node level left val Leaf)  = Node level left val (Node (level-1) Leaf x Leaf)
insertFreeOrRight _ x (Node level Leaf val right) = Node level (Node (level-1) Leaf x Leaf) val right
insertFreeOrRight _ x (Node level left val right) = Node level left val (insertFreeOrRight (level-1) x right)


Testing

*Main> foldTree "ABC"
Node 1 (Node 0 Leaf 'B' Leaf) 'C' (Node 0 Leaf 'A' Leaf)

*Main> foldTree "ABCDE"
Node 2 (Node 1 (Node 0 Leaf 'B' Leaf) 'D' Leaf) 'E' (Node 1 Leaf 'C' (Node 0 Leaf 'A' Leaf))

• This code is buggy. Maybe migrate to SO? (Source of the pretty print copied from here) Needs a quickcheck test of the property that height of the leaves do not differ by more than 1. – abuzittin gillifirca Sep 29 '14 at 12:42
• If tree were really balanced, 128 nodes should be 7-8 deep. but the output is 34-35 deep. Depth should be logarithmic, but foldTree returns linear depth. Just run foldTree [0 .. 15], you will see negative heights. – abuzittin gillifirca Sep 29 '14 at 15:03
• Again for foldTree [0 .. 15]. Output contains these nodes : (Node 1 Leaf 6 Leaf) (Node (-1) Leaf 0 Leaf) clearly height difference between these leaves is (1 - (-1)) = 2. (>1, hence unbalanced). – abuzittin gillifirca Sep 29 '14 at 15:10
• thank you, abuzittin! I'm going to try to figure it out, and then post my changes here with an edit crediting you for pointing out its failure to solve this problem. If I don't figure it out I'll migrate to SO per your suggestion. Thanks! – Kevin Meredith Sep 30 '14 at 0:49
• Pleasee don't change the code to reflect the answers, as that invalidates them. Post a new question instead. See codereview.stackexchange.com/help/on-topic – John Dvorak Oct 4 '14 at 17:32

From the design perspective, I see these possible conceptual problems:

1. The data structure doesn't help in determining if the required invariant holds or not.
2. Tree holds information that is outside of its scope - its level in the tree. This means, among other things, that a node can't be shared between multiple trees, which is something you want when manipulating trees (a new tree is just a slightly modified version of an old one, sharing most of its nodes).
3. Your design focuses only on creating a tree from a list. If you don't need any other operations, that's fine, but if you want to do more operations later, like insertion/deletion, merging trees, etc., it's going to be problematic. In particular, for some operations, when the total height of the tree must change, you'll have to update the whole tree. Most likely, it'd be possible to find an arbitrarily long sequence of operations, each taking O(n) time (where n is the number of nodes in the tree).

All these problems relate to having the node's level in Tree, which is currently sort-of unused. If instead you kept there the height of the subtree rooted at a node, you would be able to:

1. Easily check that a function keeps the required invariant when creating/manipulating nodes.
2. Each Tree value would be self-contained, depending only on its own subtree, thus can be shared within multiple trees.
3. Following from 2., you can implement the operations on trees in such a way that each takes just O(log n).

Also I'd suggest you to have a look at various balanced tree implementations (unless you want to explore it yourself, which is a good exercise).

On the other hand, if your aim is only to have a balanced tree, without the need of modifying it later, you could somewhat relax your requirement, and just require that the height of the tree is at most 1+log{2} n. Then you'd be able to create the tree incrementally, without knowing the total length of the input list: at a particular point, the algorithm tries to fill a node of height h. If it's full and there are more incoming data, it starts to build another node of height h+1. So the left subtree will be always a full binary tree.

From the syntactical point of view, it's very good that you annotate all functions with their types, the only thing I'd improve is to have a fixed line length limit to increase readability.