COMP6127 - Algorithm Design and Analysis

Week 6
Session 10
Dynamic Programming: Multistage Graph
and Travelling Salesman Problem
 The General Objectives
Multistage Graph
 Understand the concept of Multistage Graph solution using
Dynamic Programming

Travelling Salesman Problem

 Understand the concept of Travelling Salesman Problem
using Dynamic Programming.

2
 Outline

Multistage Graph
 Multistage Graph Problem
 Forward Technique
 Backward Technique

Travelling Salesman Problem

Travelling Salesman Problem
 Solving of Travelling Salesman Problem

3
COMP6127 - Algorithm Design and Analysis

MULTISTAGE GRAPH
 Dynamic Programming:
Multistage Graph

 Multistage Graph is a graph with special characteristics:

1. Directed Graph
2. Each edge has weight
3. Has only 1 source (called as s) and 1 sink (called as t)
4. Path from source to sink consists some stages V1 to Vk
5. All stages connect node in Vi to node Vi+1 where 1 ≤ i ≤ k
6. There are k stages, where k ≥ 2
7. Each path from s to t is consequence of choice k-2.

 Multistage Graph is a modeling that can be used to solve some real

problems.
 Example: choosing project to get maximum profit; including selecting
steps to perform each task.

5
 Dynamic Programming:
Multistage Graph Problem

 Multistage Graph Problem :

 Sortest path finding problem from source to sink in Multistage Graph.
 The problem is one of good implementation of Dynamic
Programming.

6
 Dynamic Programming:
Multistage Graph Problem

 Solving of Multistage Graph problem using Dynamic Programming

in shortest path from a node to another is a shortest path of
previous stage added by distance of one of an edge connects to a
stage.

 Forward Method
 Calculate distance to front (to sink)

 Backward Method
 Calculate distance to back (from source)

7
 Dynamic Programming:
Forward Method

 Analysis by calculating path from a node to sink

 Formula: cost(i,j) = min{c(j,k) + cost(i+1,k)}
 Calculation starts from nodes in stage k-2
 cost(i,j) is distance of path from node j in stage i to sink(t)
 c(j,l) is distance of path from node j to node l

8
 Dynamic Programming:
Forward Method
 cost(4,I) = c(I,L) = 7
 cost(4,J) = c(J,L) = 8
 cost(4,K) = c(K,L) = 11
 cost(3,F) = min { c(F,I) + cost(4,I) | c(F,J) + cost(4,J) }
 cost(3,F) = min { 12 + 7 | 9 + 8 } = 17
 cost(3,G) = min { c(G,I) + cost(4,I) | c(G,J) + cost(4,J) }
 cost(3,G) = min { 5 + 7 | 7 + 8 } = 12
 cost(3,H) = min { c(H,J) + cost(4,J) | c(H,K) + cost(4,K) }
 cost(3,H) = min { 10 + 8 | 8 + 11 } = 18
 cost(2,B) = min { c(B,F) + cost(3,F) | c(B,G) + cost(3,G) | c(B,H) + cost(3,H) }
 cost(2,B) = min { 4 + 17 | 8 + 12 | 11 + 18 } = 20
 cost(2,C) = min { c(C,F) + cost(3,F) | c(C,G) + cost(3,G) }
 cost(2,C) = min { 10 + 17 | 3 + 12 } = 15
 cost(2,D) = min { c(D,H) + cost(3,H) }
 cost(2,D) = min { 9 + 18 } = 27
 cost(2,E) = min { c(E,G) + cost(3,G) | c(E,H) + cost(3,H) }
 cost(2,E) = min { 6 + 12 | 12 + 18 } = 18
 cost(1,A) = min { c(A,B) + cost(2,B) | c(A,C) + cost(2,C) | c(A,D) + cost(2,D) | c(A,E) + cost(2,E) }
 cost(1,A) = min { 7 + 20 | 6 + 15 | 5 + 27 | 9 + 18 } = 21

 Shortest path is A-C-G-I-L with distance 21

9
 Dynamic Programming:
Backward Method

 Analysis by calculating path from source to a node

 Formula: bcost(i,j) = min{bcost(i–1,l) + c(l,j)}
 Calculation starts from nodes in stage 3
 bcost(i,j) is distance of path backward from source (s) to node j in
stage i
 c(j,l) is distance of path from node j to node l

10
 Dynamic Programming:
Backward Method
 bcost(2,B) = c(A,B) = 7
 bcost(2,C) = c(A,C) = 6
 bcost(2,D) = c(A,D) = 5
 bcost(2,E) = c(A,E) = 9.
 bcost(3,F) = min { c(B,F) + bcost(2,B) | c(C,F) + bcost(2,C) }
 bcost(3,F) = min { 4 + 7 | 10 + 6 } = 11
 bcost(3,G) = min { c(B,G) + bcost(2,B) | c(C,G) + bcost(2,C) | c(E,G) + bcost(2,E) }
 bcost(3,G) = min { 8 + 7 | 3 + 6 | 6 + 9 } = 9
 bcost(3,H) = min { c(B,H) + bcost(2,B) | c(D,H) + bcost(2,D) | c(E,H) + bcost(2,E) }
 bcost(3,H) = min { 11 + 7 | 9 + 5 | 12 + 9 } = 14
 bcost(4,I) = min { c(F,I) + bcost(3,F) | c(G,I) + bcost(3,G) }
 bcost(4,I) = min { 12 + 11 | 5 + 9 } = 14
 bcost(4,J) = min { c(F,J) + bcost(3,F) | c(G,J) + bcost(3,G) | c(H,J) + bcost(3,H) }
 bcost(4,J) = min { 9 + 11 | 7 + 9 | 10 + 14 } = 16
 bcost(4,K) = min { c(H,K) + bcost(3,H) }
 bcost(4,K) = min { 8 + 14 } = 22
 bcost(5,L) = min { c(I,L) + bcost(4,I) | c(J,L) + bcost(4,J) | c(K,L) + bcost(4,K) }
 bcost(5,L) = min { 7 + 14 | 8 + 16 | 11 + 22 } = 21

 Shortest path is A-C-G-I-L with distance 21

11
 Dynamic Programming:
Shortest Path in Multistage Graph

12
 Exercise
1. Find shortest path from node A to node L using Dynamic
Programming (forward method and backward method) !

13
 COMP6127 - Algorithm Design and Analysis

TRAVELLING SALESMAN PROBLEM

 Dynamic Programming:
Travelling Salesman Problem

 A salesman is selling his products to some addresses.

 Distance inter-addresses are different.
 He find a route to visit all addresses as quick as possible (shortest
route), then go back to his home.
 This problem is called as Traveling Salesman Problem (or TSP).

15
 Dynamic Programming:
Implementation of TSP

 Implementation of TSP in computer, each address is a node in

graph.
 Each path from an address to another is an edge that has weight.
 The salesman starting his travel from a node A (his house), he MUST
visit all addresses only 1 time and MUST go back to his house.

16
 Dynamic Programming:
TSP - Case

 Formula of Dynamic Programming in TSP

– p(i,L) = min[c(j,i) + p(j,L–{j})]

17
 Dynamic Programming:
Representation of TSP

 We use Cost Matrix, not Adjacency Matrix, why?

 If between 2 nodes has not an edge connection, means there
is no path between the nodes. If we use 0 (zero) value in
Adjacency Matrix, then Dynamic Programming algorithm will
detect as shortest distance.

18
 Dynamic Programming:
Formula of TSP with DP
 p(i,L) = min[c(j,i) + p(j,L–{j})]

 p(i,S) is distance of path from initial node to node i after path L.

 c(j,i) is distance from node j to node i.
 c(j,i) does not equal to c(i,j) because the graph contains 2 directed edges
with different weight.

 L-{j} is distance of path L subtracted by node j.

 Then, the shortest path for the graph is p(A,{B,C,D}) that means distance
of path from intial node A after going through node B, C, D by any
sequences (found the shortest path).

19
 Dynamic Programming:
Calculation of TSP
p(B,Ø)=c(A,B)=12
p(C,Ø)=c(A,C)=11
p(D,Ø)=c(A,D)=16
p(B,{C})=c(C,B)+p(C,Ø)=25
p(B,{D})=c(D,B)+p(D,Ø)=27
p(C,{B})=c(B,C)+p(B,Ø)=29
p(C,{D})=c(D,C)+p(D,Ø)=33
p(D,{B})=c(B,D)+p(B,Ø)=22
p(D,{C})=c(C,D)+p(C,Ø)=29
p(B,{C,D})= min[c(C,B)+p(C,{D})|c(D,B)+p(D,{C})]
= min[14+33|11+29] = 40
p(C,{B,D})= min[c(B,C)+p(B,{D})|c(D,C)+p(D,{B})]
= min[15+27|17+22] = 39
p(D,{B,C})= min[c(B,D)+p(B,{C})|c(C,D)+p(C,{B})]
= min[10+25|18+27] = 35
p(A,{B,C,D})= min[c(B,A)+p(B,{C,D}) | c(C,A)+p(C,{B,D}) | c(D,A)+p(D,{B,C})]
= min[15+40|8+39|9+35]
= 44
•Distance of the shortest path = 44

20
 Dynamic Programming:
Calculation of TSP
p(B,Ø)=c(A,B)=12
p(C,Ø)=c(A,C)=11
p(D,Ø)=c(A,D)=16
p(B,{C})=c(C,B)+p(C,Ø)=25
p(B,{D})=c(D,B)+p(D,Ø)=27
p(C,{B})=c(B,C)+p(B,Ø)=29
p(C,{D})=c(D,C)+p(D,Ø)=33
p(D,{B})=c(B,D)+p(B,Ø)=22
p(D,{C})=c(C,D)+p(C,Ø)=29
p(B,{C,D})= min[c(C,B)+p(C,{D}) | c(D,B)+p(D,{C})]
= min[14+33 | 11+29] = 40
p(C,{B,D})= min[c(B,C)+p(B,{D}) | c(D,C)+p(D,{B})]
= min[15+27 | 17+22] = 39
p(D,{B,C})= min[c(B,D)+p(B,{C}) | c(C,D)+p(C,{B})]
= min[10+25 | 18+27] = 35
p(A,{B,C,D})= min[c(B,A)+p(B,{C,D}) | c(C,A)+p(C,{B,D}) | c(D,A)+p(D,{B,C})]
= min[15+40 | 8+39 | 9+35]
= 44
•Distance of the shortest path = 44

21
 Exercise
1. Given a Cost Matrix of TSP as:

  17  8 12
10  15 11 14

 8 16  9 10
 
12 19 10  13
15 7 19   

2. Depict the TSP Graph!

3. Find the shortest path of TSP using Dynamic Programming !

22
Thank You