BACKTRACKING

Lecture 11
BACKTRACKING
BRANCH & BOUND
Chapter 10
Design & Analysis of Algorithms
Septermber 8
SLIIT
Contents
Backtracking
- Method
- Examples
- Time & space analysis of Backtracking
Branch and Bound

- Method
- Time & space analysis of Branch & Bound
- Comparison with backtracking
Septermber 8
SLIIT
Backtracking
Backtracking is a systematic way to

search for the solution to a problem
1. Define the solution space
2. Organize the space (graph or tree)
3. Use DFS recording best answer so far
Septermber 8
SLIIT
Example 0/1 Knapsack

n = 3, c = 30
w = {20, 15, 15}, p = {40, 25, 25}
Maximize the sum of profit
pi
without violating the weight constraint
wi c
Septermber 8
SLIIT
Example (cont.)
First, define the solution space
Let the binary vector {b1, ..., bn} define a
solution
If bi = 0, then object i not in knapsack
If bi = 1, then object i is in knapsack
Possible solutions = {0,0,0}, {0,0,1},
{0,1,0}, ..., {1,1,1}
Septermber 8
SLIIT
Example (cont.)
Second organize them into a tree/graph
A binary tree will do nicely...
0
Septermber 8
First Item
SLIIT
Second Item
Third Item
Example (cont.)
Now DFS, recording best solution so far
and ignoring paths that represent infeasible
solutions
Best = {0,0,0}
Profit = 0
Path = {0,0,0}
0
0
Septermber 8
SLIIT
1
7
Example (cont.)
w = {20, 15, 15} p = {40, 25, 25} c = 30
New Path = {0,0,1}
Feasible? Yes 15 < 30
Best so far = {0,0,0}

Profit = 0
Current = {0,0,1}
Profit = 25
Best = {0,0,1}
Profit = 25
0
0
25
Septermber 8
SLIIT
Example (cont.)
w = {20, 15, 15} p = {40, 25, 25} c = 30
New Path = {0,1,_}
Feasible? Yes 15 < 30.
0
0
25
Septermber 8
SLIIT
Example (cont.)
w = {20, 15, 15} p = {40, 25, 25} c = 30
Profit = 25
New Path = {0,1,0}

Feasible? Yes 15 < 30.
0
0
25 25
Septermber 8
Best = {0,0,1}
Profit = 25
Current = {0,1,0}
Profit = 25
SLIIT
1
10
Example (cont.)
w = {20, 15, 15} p = {40, 25, 25} c = 30

New Path = {0,1,1}
Feasible? Yes 30 30
Best So far = {0,0,1}

Profit = 25
0
0
25 25
Septermber 8
1
50
Current = {0,1,1}
Profit = 50
Best = {0,1,1}
Profit = 50
SLIIT
1
11
Example (cont.)
w = {20, 15, 15} p = {40, 25, 25} c = 30
New Path = {1,_,_}
Feasible? Yes 20 30
0
0
25 25
Septermber 8
50
SLIIT
12
Example (cont.)
w = {20, 15, 15} p = {40, 25, 25} c = 30
New Path = {1,0,_}
Feasible? Yes 20 30
0
0
25 25
Septermber 8
1
50
SLIIT
1
13
Example (cont.)
w = {20, 15, 15} p = {40, 25, 25} c = 30
New Path = {1,0,0}
Feasible? Yes 20 30
Best so far= {0,1,1}

Profit = 50
0
0
25 25
Septermber 8
1
50
Current = {1,0,0}
Profit = 40
Best = {0,1,1}
Profit = 50
40
SLIIT
14
Example (cont.)
w = {20, 15, 15} p = {40, 25, 25} c = 30
New Path = {1,0,1}
Feasible? No 35 not 30
0
0
25 25
Septermber 8
50 40
SLIIT
1
15
Example (cont.)
w = {20, 15, 15} p = {40, 25, 25} c = 30
New Path = {1,1,_}
Feasible? No 35 not 30

Profit = 50
0
0
25 25
Septermber 8
1
50
40
SLIIT
16
Backtracking
If it found that the newly reached node is
infeasible (I.e. the current weight is more
than the allowed weight) then there is no
point of expanding its sub trees.
Septermber 8
SLIIT
17
currentSoln {0,0,0} currentW 0 currentP 0
{0,0,0} 0 0
{0,0,0} 0 0
{1,0,0}20 40
{0,1,0}15 25
{1,0,0}20 40
{0,0,1}15 25
{1,0,0}20 40
{0,0,0}0 0
{0,1,0}15 25
Septermber 8
{0,1,1}30SLIIT
50
18
Analysis
How many nodes in the tree?
If there are n items then there are O(2n)tree
leaf nodes
Total no of nodes are (2n+1 1)
Our example, 0/1 Knapsack has O(23) leaf
nodes and (23+1 1=15) nodes
Time taken is (2n)
Note that not all might be traversed
Septermber 8
SLIIT
19
Permutation tree
In other cases the solution is a permutation
of n elements.
For example: traveling salesman
As the name suggests, the travelingsalesperson problem may be used to model
the territory covered by a salesperson.
In this problem we are given an n vertex
network and are to find a cycle of minimum
cost that includes all n vertices.
Septermber 8
SLIIT
20
Traveling salesman
1
30
2
5
4
3
20
10
Select vertex 1 as the start and end vertex.

Septermber 8
SLIIT
21
Traveling salesman
1
30
2
5
4
3
20
Septermber 8
10
Tours:
{1,2,3,4,1}
{1,2,4,3,1}
{1,3,2,4,1}
{1,3,4,2,1}
{1,4,2,3,1}
{1,4,3,2,1}
No of tours = 3! = 6
(since we are always start
with the first node,
permutations are only for the
remaining 3 nodes)
SLIIT
22
We can model the solution space for this problem using a

tree
30
1
6
B
2
3
F
4
L
1
H
4
M
1
Septermber 8
3
P
O
1
3
K
20
G
3
5
4
SLIIT
2
Q
1
W
23
Traveling salesman(contd.)
The possible tours may be described by a tree
which each roottoleaf path defines a tour. The
edgelabelsonthepathfromroottoaleafdefinea
tour. For example, the path to node R represents
thetour1,2,3,4,1.
Everytourinthenetworkisrepresentedbyexactly
one roottoleaf path in the tree. As a result, the
numberofleavesinthetreeis(n1)!
Here n = 4 and we have (4-1)! Leaf nodes
Septermber 8
SLIIT
24
A backtracking algorithm will find a minimumcost tour by searching the solution space tree in
a depth-first manner, beginning at the root.
A possible search using above Figure would
move from node A to B to C to F to L to R. At R
the tour 1,2,3,4,1 is recorded as the best tour
seen so far. Its cost is 59.
Septermber 8
SLIIT
25
A
1
F
4
L
1
H
4
M
1
3
P
O
1
3
K
20
G
3
5
4
6
B
30
Q
1
1
W
59
Septermber 8
SLIIT
26
From L we backtrack to the node C. and we
move forward to G and then to M and to S.
We have now constructed the tour 1,2,4,3,1
whose cost is 66. Since this tour isnt
superior to the best tour we have, we
discard the new tour and backtrack to G,
then , C and then B.
Septermber 8
SLIIT
27
A
1
F
4
L
1
H
4
M
1
59
66
Septermber 8
3
P
O
1
20
10
G
3
5
4
6
B
30
Q
1
SLIIT
1
W
28
Speeding up the solution

We can speed the search for an optimal
solution by determining whether or not a
newly reached node can possibly lead to a
solution better than the best found so far.
by the time we reach node I of we have
found the tour 1,3,2,4,1 with cost 25. At
node I the partial tour is 1,3,4 whose cost is
26.
There is no point in searching the subtree
with root I.
Septermber 8
SLIIT
29
A
1
2
3
F
4
L
1
M
1
S
59
66
Septermber 8
26
2
J
3
P
1
U
59
25
SLIIT
2
Q
1
T
3
K
H
4
20
G
3
5
4
6
B
30
1
W
59
30
Contents
Backtracking
- Method
- Examples
Branch and Bound

- Method
Septermber 8
SLIIT
31
Branch & Bound

searches solution tree in either a
- breadth-first or
- least-cost manner.
Septermber 8
SLIIT
32
Breadth-first
F
4
G
3
L
1
H
4
M
1
59
66
3
K
1
S
Septermber 8
20
5
4
30
3
P
O
1
2
Q
25
66
59
59
SLIIT
33
Comparison
Backtracking DFS
Branch & Bound BFS or Least cost
The memory needed by backtracking is O(length
of longest path in the solution space organization)
The memory needed by branch and bound is
O(size of solution space organization).
Branch and Bound Need more memory than

Backtracking.
Septermber 8
SLIIT
34
Summary
Backtracking
- Method
- Examples
Branch and Bound

- Method
Septermber 8
SLIIT
35

BACKTRACKING

Uploaded by

Copyright:

Available Formats

BACKTRACKING

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

BACKTRACKING

Uploaded by

Copyright:

Available Formats

Lecture 11

Branch and Bound

Backtracking is a systematic way to

Example 0/1 Knapsack

Best so far = {0,0,0}

New Path = {0,1,0}

w = {20, 15, 15} p = {40, 25, 25} c = 30

Best So far = {0,0,1}

Best so far= {0,1,1}

Best so far = {0,1,1}

currentSoln {0,0,0} currentW 0 currentP 0

Select vertex 1 as the start and end vertex.

We can model the solution space for this problem using a

Speeding up the solution

Branch and Bound

Branch & Bound

Branch and Bound Need more memory than

Branch and Bound

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.