An Introduction to Programming though C Abhiram G

An Introduction to Programming though C++ Abhiram G. Ranade Ch. 24: Structural Recursion (Part 1)

The Dictionary problem • Store a set S on a computer such that insertions and lookups can be performed quickly – S. insert(e) : insert e into the set S (and modify S). – S. find(e) : Return true iff e is a member of S. • We have seen 3 ways of storing sets – Elements stored in sorted order in an array/vector – Same, but unsorted – Elements stored Heap/Priority Queue • For all 3, either insert or find will take time O(n) i. e. proportional to the size n of the set. Goal: both operations should happen fast, i. e. in O(log n) time. • Solution: Next. – used in the Standard Library class set, and also map.

A technical assumption Elements to be stored in the set are “ordered”, i. e. they can be compared using the operator < • Numbers are “ordered” • string objects are also ordered (lexicographical order). • You will see soon how this assumption helps.

Overview • The set is represented as a binary tree – “binary search tree” • A binary search tree is a type of a graph • A binary search tree is represented on a computer in a natural way • Inserting into set, finding element – First understand as mathematical operations on the graph – Then we see procedures which operate on the computer representation.

Binary search tree A rooted tree with • At most 2 children • An element stored at each vertex – “vertex attribute” Search tree property: For all vertices v Elements stored < In left subtree Element Stored at v < Elements stored In right subtree

Binary search trees can be used to represent sets • S = {18, 34, 40, 56, 70} can be represented using many search trees. 34 56 18 56 40 18 70 70 34 40

More on binary search trees • A vertex, its children, their children, and so on are said to constitute a subtree of the vertex. • The left subtree of a vertex is the subtree of its left child; similarly right. • We can have an empty tree, i. e. a tree without even one node. This will be useful to represent empty sets. • Subtrees of a search tree are also search trees. – Recursive structure. Hence recursive algorithms!

Example Left subtree of ‘ 56’ = Subtree of ’ 18’ a BST Right subtree of ‘ 56’ = Subtree of ’ 70’ a BST. 56 18 70 60 34 40 70

Computer implementation: rooted tree • A tree node can be stored as a struct. • It will hold the value associated with it, and pointers to left and right children. struct Node{ Node *left, *right; int value; }; • If the left or the right child is absent, the corresponding pointer is NULL. • We could have used: vector<Node*> chidren;

Computer implementation: set • Set = tree • How do I refer to a tree? – Refer using the root node. – Once you are given the root node you can follow the left and right pointers to get to other nodes. . • However, with this, the empty set cannot be represented. Better: identify a set using a pointer to root of the tree holding the elements. If the set is empty, the pointer will be NULL.

A sample main program int main(){ Node* myset = NULL; insert(myset, 40); insert(myset, 20); cout <<“Finding 30: ” << find(myset, 30) << endl; // Should print 0. cout <<“Finding 40: ” << find(myset, 40) << endl; // Should print 1. }

Implementing find(myset, x) • If myset == NULL, the set is empty, so we can return false immediately. • Else we know that the values stored in the set represented by myset consist of 1. 2. 3. values in subtree of myset->left. myset->values in subtree of myset->right. • Because of the search tree property we know that the values in 1 are smaller than the value 2, which in turn is smaller than the values in 3. • If x == myset->value, then we can return true. • If x < myset->value, then x, if present, must be in myset->left. So we search that recursively. • If x > myset->value, then x, if present, must be in myset->right. So we search that recursively.

Code for find bool find(Node* myset, int x){ if(myset == NULL) return false; if(x == myset->value) return true; if(x < myset->value) return find(myset->left, x); else return find(myset->right, x); }

Specification of insert(myset, x) • Preconditions: “What holds before” – myset must point to root of a BST, say T 0. – x must be an integer. • Post condition: “What must hold after” – myset must point to the root of a BST, say T 1 – Nodes in T 1 should contain values held in nodes in T 0 and also x.

Design choice • Post conditions do not tell shape of T 1 • Our choice: Modify T 0 as little as possible – Nodes and edges of T 0 don’t change – New leaf created—at suitable position—to hold x • Other, better choices exist: Discussed later

Inserting x into an empty set • • Initially myset == NULL Create a new node, Place x there Let myset point to new node myset gets modified within insert() – Must pass myset by reference void insert(Node* &myset, int x);

Inserting x into a non-empty set • Initially myset != NULL • If x == myset->value, we can return immediately, – we dont want to insert anything twice. • If x < myset-> value, we must insert x into myset->left – Can be done recursively, because myset->left is also a search tree. Correctness: – Assume insert works on small trees, e. g. left subtree or right subtree. – Ensures call myset->left will correctly modify left subtree. – But this is what we need for the overall tree as well. – Base case: myset is empty – node must be created. • If x > myset->value, we must insert x into myset->right – Can be done recursively, because myset->left is also a search tree. – Correctness: similar.

The implementation of insert void insert(Node* &myset, int x){ if(myset == NULL){ myset = new Node; myset->left = myset->right = NULL; myset->value = x; } else{ if(x == myset->value) return; else if(x < myset->value) insert(myset->left, x); else insert(myset->right, x); } }

Our sample main program will now run int main(){ Node* myset = NULL; insert(myset, 40); insert(myset, 20); cout <<“Finding 30: ” << find(myset, 30) << endl; // Should print 0. cout <<“Finding 40: ” << find(myset, 40) << endl; // Should print 1. }

Time analysis for find • Count no. of times each statement is executed. • How to count without confusion: – when a statement is executed in the recursive call find(v, x), place a rupee on vertex v of the tree. – count the total amount of money we have placed on the tree

Time analysis for find (contd. ) • When find executes, vertices on some root to leaf path are visited, i. e. appear as first argument of find. • On each vertex, we place some fixed number of rupees, e. g. 1 for the first statement, 1 for the second, . . . • When we make the recursive call: – we must count the time spent (place some rupees) in copying the arguments and creating the activation frame of the call • For the time spent in the recursive call itself: – , the rupees will be placed on the child vertex and so on. • We will place a fixed number (independent of the tree height, number of nodes) of rupees on each vertex – The total number of rupees is O(path length). • Time = O(length of longest path). . . For worst x

Time taken by find • The number of recursive calls made by find = number of nodes visited. • Example: Find 9 in set {1, 2, 4, 5, 6, 7, 9} • Find will be called on every node. 1 2 4 5 6 7 9

Time taken by find • The number of recursive calls made by find = number of nodes visited. • Example: Find 9 in set {1, 2, 4, 5, 6, 7, 9} • Find will be called on nodes 5, 7, 9. 5 2 1 7 4 6 9

Time taken by find • Worst case: time taken = O(n), where n is the number of nodes. • Best case: time taken = O(log n). • Analysis for insert: similar. • Since we care about the worst case, this looks bad. Remark 1: • With cleverer insertion algorithm height remains small. • Time = O(logn) for insertion and find LATER Remark 2: Theorem: If nodes are inserted in random order, h will be smaller than 2 log n on the average even for simple insertion algorithm. • “Average” taken over all possible orders of inserting the elements. • Probability[time to insert n elements is > 4 nlogn] is very small. • “Average case analysis” as opposed to usual worst case.

Exercises 1. In what order should insertions be made so that the binary search tree on the left results? 2. What search tree would you get if you inserted keys in order 2, 7, 1, 5, 6, 9, 4 5 2 1 7 4 6 9

Deleting a value - 1 • Use find code to locate node v where the value is present. • If v = leaf, then v can be deleted. • Example: delete 4. • If v has 1 child, shortcut. • Example delete 2. • What v has 2 children? 5 2 1 7 4 6 9

Deleting a value - 2 • Suppose we have to delete a value x stored in a node v having 2 children. • Find node v’ holding next largest value x’ in the tree. – Leftmost node in right subtree under v. • Recursively delete x’. – Infinite recursion possible? – x’ is easy to delete. v’ has 0 or 1 children. • Replace x by x’. • Exercise: write the code.

Printing the values stored in a search tree in sorted order • Exercise void print(Node* myset){ if(myset == NULL) return; print(myset->left); cout << myset->value <<endl; print(myset->right); }

Time taken to print • Each statement executed at most once for each tree node. = Proportional to number of nodes in tree = O(n)

A new sorting algorithm Assumption: keys to be sorted are distinct. • Insert keys into a binary search tree • Then print out the values. Assuming input order is random, or additional balancing operations are done: • Time = O(nlogn) to insert + O(n) to print, so total O(nlong). If keys are not distinct: • We need to enable a multiset to be stored. • Keys in left subtree ≤ key at root ≤ Keys in right subtree • Insert procedure will have to be slightly modified.

A packaging question • When writing a program, it is nicer to declare sets by writing “Set myset; ” rather than “Node* myset; ” • We can do this by using two classes, Node as before and another Set class. • They refer to each other, and so a forward declaration is needed.

The class Set struct Node; // forward declaration class Set{ Node *p. Root; // pointer to root public: Set(){p. Root = NULL; } // empty set void insert(int x); bool find(int x); }; struct Node{ Set left, right; int value; Node(int x){value = x; } }; // left, right: initialized automatically!

Find bool Set: : find(int x){ if(p. Root == NULL) return false; if(x == p. Root->value) return true; else if(x < p. Root->value) return p. Root->left. find(x); else return p. Root->right. find(x); }

Insert void Set: : insert(int x){ if(p. Root == NULL) p. Root = new Node(x); else if(x < p. Root->value) p. Root->left. insert(x); else p. Root->right. insert(x); }

The new main program int main(){ Set myset; // constructor makes it empty myset. insert(40); myset. insert(20); cout <<“Finding 30: ” << myset. find(30) << endl; // Should print 0. cout <<“Finding 40: ” << myset. find(40) << endl; // Should print 1. }

Remarks • The new main program is nicer to read. – The user does not know that there is a tree implementing the set • The new function for insert is nicer – No references to pointers to Node. . . • The standard library class set uses the ideas we have discussed. – Also keeps tree balanced. – Time for insert, delete, find = O(longest path * time to compare) • The map class also uses these ideas + balancing. – Each node holds a (key, value) pair, s. t. keys in left subtree < key at node < keys in right subtree – Time for insert, find, delete operations = O(longest path length * time for comparing keys)

Exercises • Write a function that returns the largest value in the tree that is no larger than a given x. – If no such value, return – HUGE_VAL • Suppose each node also contains a member pointing to the parent. Assume insert function updates this correctly. Write a function which returns the value y which is smallest among all values larger than given x. – If no such value, return HUGE_VAL • Instead of a binary search tree, it is sometimes useful to use a tree with a larger degree d. In such a tree, we will store d-1 values, where ith value will be larger than all values in ith subtree. Write the code for find, insert, delete for such trees. • Suppose we implement the graph of friends using a map from strings to vectors of strings. How long would it take to print the friends of a given person? Assume string comparison takes time proportional to the length of the shorter string. • Write the code for delete.