Skip to content
Related Articles

Related Articles

Improve Article

Print Binary Search Tree in Min Max Fashion

  • Difficulty Level : Easy
  • Last Updated : 24 Jun, 2021

Given a Binary Search Tree (BST), the task is to print the BST in min-max fashion. 
What is min-max fashion? 
A min-max fashion means you have to print the maximum node first then the minimum then the second maximum then the second minimum and so on.

Examples: 

Input:                
         100                            
        /   \    
      20     500     
     /  \                      
    10   30 
          \   
           40
Output: 10 500 20 100 30 40

Input:
         40                            
        /   \    
      20     50     
     /  \      \               
    10   35     60
        /      /   
      25      55
Output: 10 60 20 55 25 50 35 40 

Approach:  

  1. Create an array inorder[] and store the inorder traversal of the givrn binary search tree.
  2. Since the inorder traversal of the binary search tree is sorted in ascending, initialise i = 0 and j = n – 1.
  3. Print inorder[i] and update i = i + 1.
  4. Print inorder[j] and update j = j – 1.
  5. Repeat steps 3 and 4 until all the elements have been printed.

Below is the implementation of the above approach:  

C++




// C++ implementation of the approach
#include <iostream>
using namespace std;
 
// Structure of each node of BST
struct node {
    int key;
    struct node *left, *right;
};
 
// A utility function to create a new BST node
node* newNode(int item)
{
    node* temp = new node();
    temp->key = item;
    temp->left = temp->right = NULL;
    return temp;
}
 
/* A utility function to insert a new
node with given key in BST */
struct node* insert(struct node* node, int key)
{
    /* If the tree is empty, return a new node */
    if (node == NULL)
        return newNode(key);
 
    /* Otherwise, recur down the tree */
    if (key < node->key)
        node->left = insert(node->left, key);
    else if (key > node->key)
        node->right = insert(node->right, key);
 
    /* return the (unchanged) node pointer */
    return node;
}
 
// Function to return the size of the tree
int sizeOfTree(node* root)
{
    if (root == NULL) {
        return 0;
    }
 
    // Calculate left size recursively
    int left = sizeOfTree(root->left);
 
    // Calculate right size recursively
    int right = sizeOfTree(root->right);
 
    // Return total size recursively
    return (left + right + 1);
}
 
// Utility function to print the
// Min max order of BST
void printMinMaxOrderUtil(node* root, int inOrder[],
                          int& index)
{
 
    // Base condition
    if (root == NULL) {
        return;
    }
 
    // Left recursive call
    printMinMaxOrderUtil(root->left, inOrder, index);
 
    // Store elements in inorder array
    inOrder[index++] = root->key;
 
    // Right recursive call
    printMinMaxOrderUtil(root->right, inOrder, index);
}
 
// Function to print the
// Min max order of BST
void printMinMaxOrder(node* root)
{
    // Store the size of BST
    int numNode = sizeOfTree(root);
 
    // Take auxiliary array for storing
    // The inorder traversal of BST
    int inOrder[numNode + 1];
    int index = 0;
 
    // Function call for printing
    // element in min max order
    printMinMaxOrderUtil(root, inOrder, index);
    int i = 0;
    index--;
 
    // While loop for printing elements
    // In front last order
    while (i < index) {
        cout << inOrder[i++] << " "
             << inOrder[index--] << " ";
    }
    if (i == index) {
        cout << inOrder[i] << endl;
    }
}
 
// Driver code
int main()
{
    struct node* root = NULL;
    root = insert(root, 50);
    insert(root, 30);
    insert(root, 20);
    insert(root, 40);
    insert(root, 70);
    insert(root, 60);
    insert(root, 80);
 
    printMinMaxOrder(root);
 
    return 0;
}

Java




// Java implementation of the approach
class GFG
{
 
// Structure of each node of BST
static class node
{
    int key;
    node left, right;
};
static int index;
 
// A utility function to create a new BST node
static node newNode(int item)
{
    node temp = new node();
    temp.key = item;
    temp.left = temp.right = null;
    return temp;
}
 
/* A utility function to insert a new
node with given key in BST */
static node insert(node node, int key)
{
    /* If the tree is empty,
    return a new node */
    if (node == null)
        return newNode(key);
 
    /* Otherwise, recur down the tree */
    if (key < node.key)
        node.left = insert(node.left, key);
    else if (key > node.key)
        node.right = insert(node.right, key);
 
    /* return the (unchanged) node pointer */
    return node;
}
 
// Function to return the size of the tree
static int sizeOfTree(node root)
{
    if (root == null)
    {
        return 0;
    }
 
    // Calculate left size recursively
    int left = sizeOfTree(root.left);
 
    // Calculate right size recursively
    int right = sizeOfTree(root.right);
 
    // Return total size recursively
    return (left + right + 1);
}
 
// Utility function to print the
// Min max order of BST
static void printMinMaxOrderUtil(node root,
                                 int inOrder[])
{
 
    // Base condition
    if (root == null)
    {
        return;
    }
 
    // Left recursive call
    printMinMaxOrderUtil(root.left, inOrder);
 
    // Store elements in inorder array
    inOrder[index++] = root.key;
 
    // Right recursive call
    printMinMaxOrderUtil(root.right, inOrder);
}
 
// Function to print the
// Min max order of BST
static void printMinMaxOrder(node root)
{
    // Store the size of BST
    int numNode = sizeOfTree(root);
 
    // Take auxiliary array for storing
    // The inorder traversal of BST
    int []inOrder = new int[numNode + 1];
 
    // Function call for printing
    // element in min max order
    printMinMaxOrderUtil(root, inOrder);
    int i = 0;
    index--;
 
    // While loop for printing elements
    // In front last order
    while (i < index)
    {
        System.out.print(inOrder[i++] + " " +
                         inOrder[index--] + " ");
    }
    if (i == index)
    {
        System.out.println(inOrder[i]);
    }
}
 
// Driver code
public static void main(String[] args)
{
    node root = null;
    root = insert(root, 50);
    insert(root, 30);
    insert(root, 20);
    insert(root, 40);
    insert(root, 70);
    insert(root, 60);
    insert(root, 80);
 
    printMinMaxOrder(root);
}
}
 
// This code is contributed by 29AjayKumar

Python3




# Python3 implementation of the approach
 
# Structure of each node of BST
class Node:
    def __init__(self,key):
        self.left = None
        self.right = None
        self.val = key
         
def insert(root,node):
    if root is None:
        root = Node(node)
    else:
        if root.val < node:
            if root.right is None:
                root.right = Node(node)
            else:
                insert(root.right, node)
        else:
            if root.left is None:
                root.left = Node(node)
            else:
                insert(root.left, node)
                 
# Function to return the size of the tree
def sizeOfTree(root):
 
    if root == None:
        return 0
     
    # Calculate left size recursively
    left = sizeOfTree(root.left)
 
    # Calculate right size recursively
    right = sizeOfTree(root.right);
 
    # Return total size recursively
    return (left + right + 1)
 
# Utility function to print the
# Min max order of BST
def printMinMaxOrderUtil(root, inOrder, index):
 
    # Base condition
    if root == None:
        return
 
    # Left recursive call
    printMinMaxOrderUtil(root.left, inOrder, index)
 
    # Store elements in inorder array
    inOrder[index[0]] = root.val
    index[0] += 1
 
    # Right recursive call
    printMinMaxOrderUtil(root.right, inOrder, index)
 
# Function to print the
# Min max order of BST
def printMinMaxOrder(root):
     
    # Store the size of BST
    numNode = sizeOfTree(root);
 
    # Take auxiliary array for storing
    # The inorder traversal of BST
    inOrder = [0] * (numNode + 1)
    index = 0
 
    # Function call for printing
    # element in min max order
    ref = [index]
    printMinMaxOrderUtil(root, inOrder, ref)
    index = ref[0]
    i = 0;
    index -= 1
 
    # While loop for printing elements
    # In front last order
    while (i < index):
         
        print (inOrder[i], inOrder[index], end = ' ')
        i += 1
        index -= 1
     
    if i == index:
        print(inOrder[i])
     
# Driver Code
root = Node(50)
insert(root, 30)
insert(root, 20)
insert(root, 40)
insert(root, 70)
insert(root, 60)
insert(root, 80)
 
printMinMaxOrder(root)
     
# This code is contributed by Sadik Ali

C#




// C# implementation of the approach
using System;
 
class GFG
{
 
// Structure of each node of BST
class node
{
    public int key;
    public node left, right;
};
static int index;
 
// A utility function to create a new BST node
static node newNode(int item)
{
    node temp = new node();
    temp.key = item;
    temp.left = temp.right = null;
    return temp;
}
 
/* A utility function to insert a new
node with given key in BST */
static node insert(node node, int key)
{
    /* If the tree is empty,
    return a new node */
    if (node == null)
        return newNode(key);
 
    /* Otherwise, recur down the tree */
    if (key < node.key)
        node.left = insert(node.left, key);
    else if (key > node.key)
        node.right = insert(node.right, key);
 
    /* return the (unchanged) node pointer */
    return node;
}
 
// Function to return the size of the tree
static int sizeOfTree(node root)
{
    if (root == null)
    {
        return 0;
    }
 
    // Calculate left size recursively
    int left = sizeOfTree(root.left);
 
    // Calculate right size recursively
    int right = sizeOfTree(root.right);
 
    // Return total size recursively
    return (left + right + 1);
}
 
// Utility function to print the
// Min max order of BST
static void printMinMaxOrderUtil(node root,
                                  int []inOrder)
{
 
    // Base condition
    if (root == null)
    {
        return;
    }
 
    // Left recursive call
    printMinMaxOrderUtil(root.left, inOrder);
 
    // Store elements in inorder array
    inOrder[index++] = root.key;
 
    // Right recursive call
    printMinMaxOrderUtil(root.right, inOrder);
}
 
// Function to print the
// Min max order of BST
static void printMinMaxOrder(node root)
{
    // Store the size of BST
    int numNode = sizeOfTree(root);
 
    // Take auxiliary array for storing
    // The inorder traversal of BST
    int []inOrder = new int[numNode + 1];
 
    // Function call for printing
    // element in min max order
    printMinMaxOrderUtil(root, inOrder);
    int i = 0;
    index--;
 
    // While loop for printing elements
    // In front last order
    while (i < index)
    {
        Console.Write(inOrder[i++] + " " +
                      inOrder[index--] + " ");
    }
    if (i == index)
    {
        Console.WriteLine(inOrder[i]);
    }
}
 
// Driver code
public static void Main(String[] args)
{
    node root = null;
    root = insert(root, 50);
    insert(root, 30);
    insert(root, 20);
    insert(root, 40);
    insert(root, 70);
    insert(root, 60);
    insert(root, 80);
 
    printMinMaxOrder(root);
}
}
 
// This code is contributed by PrinciRaj1992

Javascript




<script>
 
// Javascript implementation of the approach
 
// Structure of each node of BST
class node
{
    constructor()
    {
        this.key = 0;
        this.left = null;
        this.right = null;
    }
};
 
var index = 0;
 
// A utility function to create a new BST node
function newNode(item)
{
    var temp = new node();
    temp.key = item;
    temp.left = temp.right = null;
    return temp;
}
 
// A utility function to insert a new
// node with given key in BST
function insert(node, key)
{
     
    /* If the tree is empty,
    return a new node */
    if (node == null)
        return newNode(key);
 
    /* Otherwise, recur down the tree */
    if (key < node.key)
        node.left = insert(node.left, key);
    else if (key > node.key)
        node.right = insert(node.right, key);
 
    /* return the (unchanged) node pointer */
    return node;
}
 
// Function to return the size of the tree
function sizeOfTree(root)
{
    if (root == null)
    {
        return 0;
    }
 
    // Calculate left size recursively
    var left = sizeOfTree(root.left);
 
    // Calculate right size recursively
    var right = sizeOfTree(root.right);
 
    // Return total size recursively
    return (left + right + 1);
}
 
// Utility function to print the
// Min max order of BST
function printMinMaxOrderUtil(root, inOrder)
{
     
    // Base condition
    if (root == null)
    {
        return;
    }
 
    // Left recursive call
    printMinMaxOrderUtil(root.left, inOrder);
 
    // Store elements in inorder array
    inOrder[index++] = root.key;
 
    // Right recursive call
    printMinMaxOrderUtil(root.right, inOrder);
}
 
// Function to print the
// Min max order of BST
function printMinMaxOrder(root)
{
     
    // Store the size of BST
    var numNode = sizeOfTree(root);
 
    // Take auxiliary array for storing
    // The inorder traversal of BST
    var inOrder = Array(numNode+1);
 
    // Function call for printing
    // element in min max order
    printMinMaxOrderUtil(root, inOrder);
    var i = 0;
    index--;
 
    // While loop for printing elements
    // In front last order
    while (i < index)
    {
        document.write(inOrder[i++] + " " +
                       inOrder[index--] + " ");
    }
    if (i == index)
    {
        document.write(inOrder[i]);
    }
}
 
// Driver code
var root = null;
root = insert(root, 50);
insert(root, 30);
insert(root, 20);
insert(root, 40);
insert(root, 70);
insert(root, 60);
insert(root, 80);
 
printMinMaxOrder(root);
 
// This code is contributed by noob2000
 
</script>
Output: 
20 80 30 70 40 60 50

 

Attention reader! Don’t stop learning now. Get hold of all the important DSA concepts with the DSA Self Paced Course at a student-friendly price and become industry ready.  To complete your preparation from learning a language to DS Algo and many more,  please refer Complete Interview Preparation Course.

In case you wish to attend live classes with experts, please refer DSA Live Classes for Working Professionals and Competitive Programming Live for Students.




My Personal Notes arrow_drop_up
Recommended Articles
Page :