0/1 Knapsack using Least Cost Branch and Bound

Given N items with weights W[0..n-1], values V[0..n-1] and a knapsack with capacity C, select the items such that: 

1. The sum of weights taken into the knapsack is less than or equal to C.
2. The sum of values of the items in the knapsack is maximum among all the possible combinations.

Examples:

Input: N = 4, C = 15, V[]= {10, 10, 12, 18}, W[]= {2, 4, 6, 9} 
Output:
Items taken into the knapsack are 
1 1 0 1 
Maximum profit is 38 
Explanation:
1 in the output indicates that the item is included in the knapsack while 0 indicates that the item is excluded. 
Since the maximum possible cost allowed is 15, the ways to select items are: 
(1 1 0 1) -> Cost = 2 + 4 + 9 = 15, Profit = 10 + 10 + 18 = 38. 
(0 0 1 1) -> Cost = 6 + 9 = 15, Profit = 12 + 18 = 30 
(1 1 1 0) -> Cost = 2 + 4 + 6 = 12, Profit = 32 
Hence, maximum profit possible within a cost of 15 is 38.
Input: N = 4, C = 21, V[]= {18, 20, 14, 18}, W[]= {6, 3, 5, 9} 
Output:
Items taken into the knapsack are 
1 1 0 1 
Maximum profit is 56 
Explanation:
Cost = 6 + 3 + 9 = 18 
Profit = 18 + 20 + 18 = 56 

Approach:
In this post, the implementation of Branch and Bound method using Least cost(LC) for 0/1 Knapsack Problem is discussed.
Branch and Bound can be solved using FIFO, LIFO and LC strategies. The least cost(LC) is considered the most intelligent as it selects the next node based on a Heuristic Cost Function. It picks the one with the least cost. 
As 0/1 Knapsack is about maximizing the total value, we cannot directly use the LC Branch and Bound technique to solve this. Instead, we convert this into a minimization problem by taking negative of the given values. 
Follow the steps below to solve the problem: 

1. Sort the items based on their value/weight(V/W) ratio.
2. Insert a dummy node into the priority queue.
3. Repeat the following steps until the priority queue is empty:
   • Extract the peek element from the priority queue and assign it to the current node.
   • If the upper bound of the current node is less than minLB, the minimum lower bound of all the nodes explored, then there is no point of exploration. So, continue with the next element. The reason for not considering the nodes whose upper bound is greater than minLB is that, the upper bound stores the best value that might be achieved. If the best value itself is not optimal than minLB, then exploring that path is of no use. 
     
   • Update the path array.
   • If the current node’s level is N, then check whether the lower bound of the current node is less than finalLB, minimum lower bound of all the paths that reached the final level. If it is true, update the finalPath and finalLB. Otherwise, continue with the next element.
   • Calculate the lower and upper bounds of the right child of the current node.
   • If the current item can be inserted into the knapsack, then calculate the lower and upper bound of the left child of the current node.
   • Update the minLB and insert the children if their upper bound is less than minLB.

Illustration:
N = 4, C = 15, V[]= {10 10 12 18}, W[]= {2 4 6 9} 

Left branch and right branch at i^th level stores the maximum obtained including and excluding the i^th element.
Below image shows the state of the priority queue after every step: 

Below is the implementation of the above approach:

Java

// Java Program to implement
// 0/1 knapsack using LC
// Branch and Bound

import java.util.*;
class Item {

    // Stores the weight
    // of items
    float weight;

    // Stores the values
    // of items
    int value;

    // Stores the index
    // of items
    int idx;
    public Item() {}
    public Item(int value, float weight,
                int idx)
    {
        this.value = value;
        this.weight = weight;
        this.idx = idx;
    }
}

class Node {
    // Upper Bound: Best case
    // (Fractional Knapsack)
    float ub;

    // Lower Bound: Worst case
    // (0/1)
    float lb;

    // Level of the node in
    // the decision tree
    int level;

    // Stores if the current
    // item is selected or not
    boolean flag;

    // Total Value: Stores the
    // sum of the values of the
    // items included
    float tv;

    // Total Weight: Stores the sum of
    // the weights of included items
    float tw;
    public Node() {}
    public Node(Node cpy)
    {
        this.tv = cpy.tv;
        this.tw = cpy.tw;
        this.ub = cpy.ub;
        this.lb = cpy.lb;
        this.level = cpy.level;
        this.flag = cpy.flag;
    }
}

// Comparator to sort based on lower bound
class sortByC implements Comparator<Node> {
    public int compare(Node a, Node b)
    {
        boolean temp = a.lb > b.lb;
        return temp ? 1 : -1;
    }
}

class sortByRatio implements Comparator<Item> {
    public int compare(Item a, Item b)
    {
        boolean temp = (float)a.value
                           / a.weight
                       > (float)b.value
                             / b.weight;
        return temp ? -1 : 1;
    }
}

class knapsack {

    private static int size;
    private static float capacity;

    // Function to calculate upper bound
    // (includes fractional part of the items)
    static float upperBound(float tv, float tw,
                            int idx, Item arr[])
    {
        float value = tv;
        float weight = tw;
        for (int i = idx; i < size; i++) {
            if (weight + arr[i].weight
                <= capacity) {
                weight += arr[i].weight;
                value -= arr[i].value;
            }
            else {
                value -= (float)(capacity
                                 - weight)
                         / arr[i].weight
                         * arr[i].value;
                break;
            }
        }
        return value;
    }

    // Calculate lower bound (doesn't
    // include fractional part of items)
    static float lowerBound(float tv, float tw,
                            int idx, Item arr[])
    {
        float value = tv;
        float weight = tw;
        for (int i = idx; i < size; i++) {
            if (weight + arr[i].weight
                <= capacity) {
                weight += arr[i].weight;
                value -= arr[i].value;
            }
            else {
                break;
            }
        }
        return value;
    }

    static void assign(Node a, float ub, float lb,
                       int level, boolean flag,
                       float tv, float tw)
    {
        a.ub = ub;
        a.lb = lb;
        a.level = level;
        a.flag = flag;
        a.tv = tv;
        a.tw = tw;
    }

    public static void solve(Item arr[])
    {
        // Sort the items based on the
        // profit/weight ratio
        Arrays.sort(arr, new sortByRatio());

        Node current, left, right;
        current = new Node();
        left = new Node();
        right = new Node();

        // min_lb -> Minimum lower bound
        // of all the nodes explored

        // final_lb -> Minimum lower bound
        // of all the paths that reached
        // the final level
        float minLB = 0, finalLB
                         = Integer.MAX_VALUE;
        current.tv = current.tw = current.ub
            = current.lb = 0;
        current.level = 0;
        current.flag = false;

        // Priority queue to store elements
        // based on lower bounds
        PriorityQueue<Node> pq
            = new PriorityQueue<Node>(
                new sortByC());

        // Insert a dummy node
        pq.add(current);

        // curr_path -> Boolean array to store
        // at every index if the element is
        // included or not

        // final_path -> Boolean array to store
        // the result of selection array when
        // it reached the last level
        boolean currPath[] = new boolean[size];
        boolean finalPath[] = new boolean[size];

        while (!pq.isEmpty()) {
            current = pq.poll();
            if (current.ub > minLB
                || current.ub >= finalLB) {
                // if the current node's best case
                // value is not optimal than minLB,
                // then there is no reason to
                // explore that node. Including
                // finalLB eliminates all those
                // paths whose best values is equal
                // to the finalLB
                continue;
            }

            if (current.level != 0)
                currPath[current.level - 1]
                    = current.flag;

            if (current.level == size) {
                if (current.lb < finalLB) {
                    // Reached last level
                    for (int i = 0; i < size; i++)
                        finalPath[arr[i].idx]
                            = currPath[i];
                    finalLB = current.lb;
                }
                continue;
            }

            int level = current.level;

            // right node -> Excludes current item
            // Hence, cp, cw will obtain the value
            // of that of parent
            assign(right, upperBound(current.tv,
                                     current.tw,
                                     level + 1, arr),
                   lowerBound(current.tv, current.tw,
                              level + 1, arr),
                   level + 1, false,
                   current.tv, current.tw);

            if (current.tw + arr[current.level].weight
                <= capacity) {

                // left node -> includes current item
                // c and lb should be calculated
                // including the current item.
                left.ub = upperBound(
                    current.tv
                        - arr[level].value,
                    current.tw
                        + arr[level].weight,
                    level + 1, arr);
                left.lb = lowerBound(
                    current.tv
                        - arr[level].value,
                    current.tw
                        + arr[level].weight,
                    level + 1,
                    arr);
                assign(left, left.ub, left.lb,
                       level + 1, true,
                       current.tv - arr[level].value,
                       current.tw
                           + arr[level].weight);
            }

            // If the left node cannot
            // be inserted
            else {

                // Stop the left node from
                // getting added to the
                // priority queue
                left.ub = left.lb = 1;
            }

            // Update minLB
            minLB = Math.min(minLB, left.lb);
            minLB = Math.min(minLB, right.lb);

            if (minLB >= left.ub)
                pq.add(new Node(left));
            if (minLB >= right.ub)
                pq.add(new Node(right));
        }
        System.out.println("Items taken"
                           + "into the knapsack are");
        for (int i = 0; i < size; i++) {
            if (finalPath[i])
                System.out.print("1 ");
            else
                System.out.print("0 ");
        }
        System.out.println("\nMaximum profit"
                           + " is " + (-finalLB));
    }

    // Driver code
    public static void main(String args[])
    {
        size = 4;
        capacity = 15;

        Item arr[] = new Item[size];
        arr[0] = new Item(10, 2, 0);
        arr[1] = new Item(10, 4, 1);
        arr[2] = new Item(12, 6, 2);
        arr[3] = new Item(18, 9, 3);

        solve(arr);
    }
}

Output

Items taken into the knapsack are : 
1 1 0 1 
Maximum profit is : 38