Implement heap increasePriority and let Dijkstra use it

Author: caojoshua

data_structures/heap/heap.ts

@@ -12,16 +12,13 @@
  */
 
 export abstract class Heap<T> {
-  private heap: T[];
+  protected heap: T[];
   // A comparison function. Returns true if a should be the parent of b.
-  private compare: (a: any, b: any) => boolean;
+  private compare: (a: T, b: T) => boolean;
 
-  constructor(elements: T[] = [], compare: (a: T, b: T) => boolean) {
+  constructor(compare: (a: T, b: T) => boolean) {
     this.heap = [];
     this.compare = compare;
-    for (let element of elements) {
-      this.insert(element);
-    }
   }
 
   /**
@@ -68,17 +65,20 @@ export abstract class Heap<T> {
     return this.size() === 0;
   }
 
-  private bubbleUp(): void {
-    let index = this.size() - 1;
+  protected swap(a: number, b: number) {
+    [this.heap[a], this.heap[b]] = [
+      this.heap[b],
+      this.heap[a],
+    ];
+  }
+
+  protected bubbleUp(index = this.size() - 1): void {
     let parentIndex;
 
     while (index > 0) {
       parentIndex = Math.floor((index - 1) / 2);
       if (this.isRightlyPlaced(index, parentIndex)) break;
-      [this.heap[parentIndex], this.heap[index]] = [
-        this.heap[index],
-        this.heap[parentIndex],
-      ];
+      this.swap(parentIndex, index);
       index = parentIndex;
     }
   }
@@ -95,10 +95,7 @@ export abstract class Heap<T> {
         rightChildIndex
       );
       if (this.isRightlyPlaced(childIndexToSwap, index)) break;
-      [this.heap[childIndexToSwap], this.heap[index]] = [
-        this.heap[index],
-        this.heap[childIndexToSwap],
-      ];
+      this.swap(childIndexToSwap, index);
       index = childIndexToSwap;
       leftChildIndex = this.getLeftChildIndex(index);
       rightChildIndex = this.getRightChildIndex(index);
@@ -140,13 +137,60 @@ export abstract class Heap<T> {
 }
 
 export class MinHeap<T> extends Heap<T> {
-  constructor(elements: T[] = [], compare = (a: T, b: T) => { return a < b }) {
-    super(elements, compare);
+  constructor(compare = (a: T, b: T) => { return a < b }) {
+    super(compare);
   }
 }
 
 export class MaxHeap<T> extends Heap<T> {
-  constructor(elements: T[] = [], compare = (a: T, b: T) => { return a > b }) {
-    super(elements, compare);
+  constructor(compare = (a: T, b: T) => { return a > b }) {
+    super(compare);
+  }
+}
+
+// Priority queue that supports increasePriority() in O(log(n)). The limitation is that there can only be a single element for each key, and the max number or keys must be specified at heap construction. Most of the functions are wrappers around MinHeap functions and update the keys array.
+export class PriorityQueue<T> extends MinHeap<T> {
+  // Maps from the n'th node to its index within the heap.
+  private keys: number[];
+  // Maps from element to its index with keys.
+  private keys_index: (a: T) => number;
+
+  constructor(keys_index: (a: T) => number, num_keys: number, compare = (a: T, b: T) => { return a < b }) {
+    super(compare);
+    this.keys = Array(num_keys).fill(-1);
+    this.keys_index = keys_index;
+  }
+
+  protected swap(a: number, b: number) {
+    let akey = this.keys_index(this.heap[a]);
+    let bkey = this.keys_index(this.heap[b]);
+    [this.keys[akey], this.keys[bkey]] = [this.keys[bkey], this.keys[akey]];
+    super.swap(a, b);
+  }
+
+  public insert(value: T) {
+    this.keys[this.keys_index(value)] = this.size();
+    super.insert(value);
+  }
+
+  public extract(): T {
+    // Unmark the the highest priority element and set key to zero for the last element in the heap.
+    this.keys[this.keys_index(this.heap[0])] = -1;
+    if (this.size() > 1) {
+      this.keys[this.keys_index(this.heap[this.size() - 1])] = 0;
+    }
+    return super.extract();
+  }
+
+  public increasePriority(idx: number, value: T) {
+    if (this.keys[idx] == -1) {
+      // If the key does not exist, insert the value.
+      this.insert(value);
+      return;
+    }
+    let key = this.keys[idx];
+    // Increase the priority and bubble it up the heap.
+    this.heap[key] = value;
+    this.bubbleUp(key);
   }
 }

data_structures/heap/test/max_heap.test.ts

@@ -7,7 +7,10 @@ describe("MaxHeap", () => {
   ];
 
   beforeEach(() => {
-    heap = new MaxHeap(elements);
+    heap = new MaxHeap();
+    for (let element of elements) {
+      heap.insert(element);
+    }
   });
 
   it("should initialize a heap from input array", () => {

data_structures/heap/test/min_heap.test.ts

@@ -1,4 +1,4 @@
-import { MinHeap } from "../heap";
+import { MinHeap, PriorityQueue } from "../heap";
 
 describe("MinHeap", () => {
   let heap: MinHeap<number>;
@@ -7,7 +7,10 @@ describe("MinHeap", () => {
   ];
 
   beforeEach(() => {
-    heap = new MinHeap(elements);
+    heap = new MinHeap();
+    for (let element of elements) {
+      heap.insert(element);
+    }
   });
 
   it("should initialize a heap from input array", () => {
@@ -27,7 +30,7 @@ describe("MinHeap", () => {
     heap.check();
   });
 
-  const extract_all = (heap: MinHeap<number>) => {
+  const extract_all = (heap: MinHeap<number>, elements: number[]) => {
     [...elements].sort((a, b) => a - b).forEach((element: number) => {
       expect(heap.extract()).toEqual(element);
     });
@@ -36,7 +39,7 @@ describe("MinHeap", () => {
   }
 
   it("should remove and return the min elements in order", () => {
-    extract_all(heap);
+    extract_all(heap, elements);
   });
 
   it("should insert all, then remove and return the min elements in order", () => {
@@ -46,6 +49,27 @@ describe("MinHeap", () => {
     });
     heap.check();
     expect(heap.size()).toEqual(elements.length);
-    extract_all(heap);
+    extract_all(heap, elements);
+  });
+
+  it("should increase priority", () => {
+    let heap = new PriorityQueue((a: number) => { return a; }, elements.length);
+    elements.forEach((element: number) => {
+      heap.insert(element);
+    });
+    heap.check();
+    expect(heap.size()).toEqual(elements.length);
+
+    heap.increasePriority(55, 14);
+    heap.increasePriority(18, 16);
+    heap.increasePriority(81, 72);
+    heap.increasePriority(9, 0);
+    heap.increasePriority(43, 33);
+    heap.check();
+    // Elements after increasing priority
+    const newElements: number[] = [
+      12, 4, 33, 42, 0, 7, 39, 16, 14, 1, 51, 34, 72, 16,
+    ];
+    extract_all(heap, newElements);
   });
 });

graph/dijkstra.ts

@@ -1,4 +1,4 @@
-import { MinHeap } from '../data_structures/heap/min_heap';
+import { MinHeap, PriorityQueue } from '../data_structures/heap/heap';
 /**
  * @function dijkstra
  * @description Compute the shortest path from a source node to all other nodes. The input graph is in adjacency list form. It is a multidimensional array of edges. graph[i] holds the edges for the i'th node. Each edge is a 2-tuple where the 0'th item is the destination node, and the 1'th item is the edge weight.
@@ -11,22 +11,24 @@ import { MinHeap } from '../data_structures/heap/min_heap';
  * @see https://en.wikipedia.org/wiki/Dijkstra%27s_algorithm
  */
 export const dijkstra = (graph: [number, number][][], start: number): number[] => {
-  // We use a priority queue to make sure we always visit the closest node. By using a MinHeap,
-  // finding the next node is O(log(V))
-  let priorityQueue = new MinHeap([start]);
+  // We use a priority queue to make sure we always visit the closest node. The
+  // queue makes comparisons based on path weights.
+  let priorityQueue = new PriorityQueue((a: [number, number]) => { return a[0] }, graph.length, (a: [number, number], b: [number, number]) => { return a[1] < b[1] });
+  priorityQueue.insert([start, 0]);
   // We save the shortest distance to each node in `distances`. If a node is
   // unreachable from the start node, its distance is Infinity.
   let distances = Array(graph.length).fill(Infinity);
   distances[start] = 0;
 
   while (priorityQueue.size() > 0) {
-    const node = priorityQueue.extract();
+    const [node, _] = priorityQueue.extract();
     graph[node].forEach(([child, weight]) => {
       let new_distance = distances[node] + weight;
       if (new_distance < distances[child]) {
         // Found a new shortest path to child node. Record its distance and add child to the queue.
+        // If the child already exists in the queue, the priority will be updated. This will make sure the queue will be at most size V (number of vertices).
+        priorityQueue.increasePriority(child, [child, weight]);
         distances[child] = new_distance;
-        priorityQueue.insert(child);
       }
     });
   }