Dynamic

Dynamic Programming
Programming
(optimization
(optimization problem)
problem)
CS3024
Lecture 10
 Outline
• Optimization problem (DP vs Greedy),
e.g making change, knapsack.
• Exercises.

21/4/06 RMB/lecture 9 2
What is dynamic programming ?

• The word programming stands for planning

not for computer programming.

• The word dynamic is related with the

manner in which the tables used in
obtaining the solution are constructed

21/4/06 RMB/lecture 9 3
What is dynamic programming ?

Dynamic programming is also related

to greedy strategy since both of
them can be applied to optimization
problems which have the property of
optimal substructure

21/4/06 RMB/lecture 9 4
 The steps in the development
of a dynamic programming
1. Establish a recursive property that
gives the optimal solution to an
instance of the problem.
2. Compute the value of an optimal
solution in a bottom-up fashion.
3. Construct an optimal solution in a
bottom-up fashion.

21/4/06 RMB/lecture 9 5
The principle of optimality

It is said to apply in a problem if an

optimal solution to an instance of a
problem always contains optimal
solution to all subinstances.

21/4/06 RMB/lecture 9 6
The principle of optimality

Dynamic programming may be used

only when the principle of optimality
holds.

21/4/06 RMB/lecture 9 7
 Application: discrete knapsack
Hypothesis:
the capacity C and the dimensions d1,…,dn are natural numbers

The problem can be reformulated as:

find (s1,s2,…,sn) with si in {0,1} such that:
s1d1 +…+ sndn <= C (constraint)
s1p1 +…+ snpn is maximal (optimization criterion)

Remark
the greedy technique can be applied but it does not guarantee the
optimality

21/4/06 RMB/lecture 9 8
 Application: discrete knapsack
Example: n=3, Greedy idea:
C=5, • Sort decreasingly the set of
d1=1, d2=2, d3=3 objects on the relative profit (pi/di)
p1=6, p2=10, p3=12 • Select the elements until the
knapsack is overloaded
Relative profit:
Greedy solution: (1,1,0)
pr1=6, pr2=5, pr3=4
Total value: V=16

Remark: this is not the optimal solution; the solution (0,1,1) is better
since V=22

21/4/06 RMB/lecture 9 9
 Application: discrete knapsack
1. Analyzing the structure of an optimal solution
Let P(i,j) be the generic problem of selecting from the set of objects {o1,
…,oi} in order to fill in a knapsack of capacity j.

Remarks:
• P(n,C) is the initial problem
• If i<n, j<C then P(i,j) is a subproblem of P(n,C)
• Let s(i,j) be an optimal solution of P(i,j). There are two situations:
– si=1 (the object oi is selected) => this lead us to the subproblem P(i-1,j-
di) and if s(i,j) is optimal then s(i-1,j-di) should be optimal
– si=0 (the object oi is not selected) => this lead us to the subproblem P(i-
1,j) and if s(i,j) is optimal then s(i-1,j) should be optimal
Thus the solution s has the optimal substructure property

21/4/06 RMB/lecture 9 10
 Application: discrete knapsack

2. Find a recurrence relation

Let V(i,j) be the total value corresponding to an optimal solution of P(i,j)

0 if i=0 or j=0 (the set is empty or the

knapsack has not capacity at all)

V(i,j) = V(i-1,j) if di>j or V(i-1,j)>V(i-1,j-di)+ pi

(either the object i doesn’t fit the knapsack or by
selecting it we obtain a worse solution than by not
selecting it)

V(i-1,j-di)+pi otherwise

21/4/06 RMB/lecture 9 11
 Application: discrete knapsack
The recurrence relation can be written also as:

0 if i=0 or j=0

V(i,j) = V(i-1,j) if di>j

max{V(i-1,j), V(i-1,j-di)+ pi } if di<=j

Remarks:
• for the problem P(n,C) the table V has (n+1) rows and (C+1)
columns
• V(n,C) gives us the value corresponding to the optimal solution

21/4/06 RMB/lecture 9 12
 Application: discrete knapsack
Example: V

0 if i=0 or j=0 0 1 2 3 4 5
0 0 0 0 0 0 0
V(i,j) = V(i-1,j) if d i>j
1 0 6 6 6 6 6
max{V(i-1,j),
V(i-1,j-di)+ pi } if di<=j 2 0 6 10 16 16 16

3 0 6 10 16 18 22
d: 1 2 3
p: 6 10 12

21/4/06 RMB/lecture 9 13
 Application: discrete knapsack
3. Developing the recurrence Algorithm:
relation
computeV (p[1..n],d[1..n],C)
FOR i:=0,n DO V[i,0]:=0
0 if i=0 or j=0
FOR j:=1,n DO V[0,j]:=0
FOR i:=1,n DO
V(i,j) = V(i-1,j) if d i>j
FOR j:=1,C DO
IF j<d[i] THEN V[i,j]:=V[i-1,j]
max{V(i-1,j),
ELSE
V(i-1,j-di)+ pi } if di<=j
V[i,j]:=max(V[i-1,j],V[i-1,j-d[i]]+p[i])
RETURN V[0..n,0..C]

21/4/06 RMB/lecture 9 14
 Application: discrete knapsack
4. Constructing the solution Steps:

Example: • Compare V[3,5] with V[2,5]. Since

they are different it means that the
object o3 is selected
0 1 2 3 4 5
• Go to V[2,5-d3]=V[2,2]=10 and
0 0 0 0 0 0 0 compare it with V[1,2]=6. Since
they are different it means that also
1 0 6 6 6 6 6 o2 is selected
• Go to V[1,2-d2]=V[1,0]=0. Since the
2 0 6 10 16 16 16 current capacity is 0 we cannot
select another object

3 0 6 10 16 18 22
Thus the solution is {o2,o3} or s=(0,1,1)

21/4/06 RMB/lecture 9 15
 Application: discrete knapsack
4. Constructing the solution Algorithm:

Example: Construct(V[0..n,0..C],d[1..n])
FOR i:=1,n DO s[i]:=0
0 1 2 3 4 5 i:=n; j:=C
0 0 0 0 0 0 0 WHILE j>0 DO
WHILE (i>1) AND (V[i,j]=V[i-1,j])
1 0 6 6 6 6 6 DO i:=i-1
s[i]:=1
2 0 6 10 16 16 16 j:=j-d[i]
i:=i-1
3 0 6 10 16 18 22 RETURN s[1..n]

21/4/06 RMB/lecture 9 16
 Problem 1
Coin changing problem

By using coins of values {v1,v2,…,vn} we want to find a subset of

coins that total the amount C and the number of used coins
is minimal

Let us suppose that vn > vn-1 > … > v1 =1

Example:
v1=1 v2=6 v3=10
C=12

Greedy solution: 10+1+1 (3 coins) Not optimal !

Better solution: 6+6 (2 coins)

21/4/06 RMB/lecture 9 17
 Problem 1
• The generic problem P(i,j) asks for finding a minimal subset
of {v1,v2,…,vi} which covers the value j
An optimal solution of P(i,j) belongs to one of the classes:
– It doesn’t use the coin vi (it is identical to the optimal solution
of P(i-1,j))
– It uses the coin vi (it contains an optimal solution of P(i,j-vi))
• Let R(i,j) be the number of coins corresponding to the
optimal solution of P(i,j)
j if i=1
R(i,j)= 0 if j=0
R(i-1,j) if vi>j
min{R(i-1,j),1+R(i,j- vi )} otherwise

• Let Q(i,j) in {0,1} be an indicator of the using of coin i in

covering the value j

21/4/06 RMB/lecture 9 18
 Problem 1
j if i=1
R(i,j)= 0 if j=0
R(i-1,j) if vi>j
min{R(i-1,j),1+R(i,j- vi )} if vi<=j

Example: C=12, v: 10, 6, 1

0 1 2 3 4 5 6 7 8 9 10 11 12

1 0 1 2 3 4 5 6 7 8 9 10 11 12

2 0 1 2 3 4 5 1 2 3 4 5 6 2

3 0 1 2 3 4 5 1 2 3 4 1 2 2

21/4/06 RMB/lecture 9 19
 Problem 1
0 if coin i is not used in the optimal solution of P(i,j)
Q(i,j)=
1 if coin i is used in the optimal solution of P(i,j)

Example:
0 1 2 3 4 5 6 7 8 9 10 11 12

1 0 1 1 1 1 1 1 1 1 1 1 1 1

2 0 0 0 0 0 0 1 1 1 1 1 1 1

3 0 0 0 0 0 0 0 0 0 0 1 1 0

21/4/06 RMB/lecture 9 20
 Problem 1 Solution
0 1 2 3 4 5 6 7 8 9 10 11 12 construction

1 0 1 2 3 4 5 6 7 8 9 10 1112
Find the minimal
2 0 1 2 3 4 5 1 2 3 4 5 6 2 value on the
last column
of R matrix
3 0 1 2 3 4 5 1 2 3 4 1 2 2 which
corresponds
to a one in Q
0 1 2 3 4 5 6 7 8 9 10 11 12 matrix
Go on the same line
1 0 1 1 1 1 1 1 1 1 1 1 1 1 on the
column
placed vi
2 0 0 0 0 0 0 1 1 1 1 1 1 1 positions to
the left and
3 0 0 0 0 0 0 0 0 0 0 1 1 0 check if the
correspondin
g value in Q
The first value equal to one starting from bottom is one …

21/4/06 RMB/lecture 9 21
 Problem 1
Construct(R[1..n,0..C],Q[1..n,0..C])
FOR i:=1,n DO S[i]:=0
imin=n
WHILE Q[imin,C]=0 DO imin:=imin-1
i:=imin
j:=C
WHILE j>0 AND i>0 DO
IF Q[i,j]=1 THEN S[i]:=S[i]+1
j:=j-v[i]
ELSE i:=i-1
IF j=0 THEN RETURN S[1..n]
ELSE “ it doesn’t exist a solution”

This can happen only if v[1]<>1

21/4/06 RMB/lecture 9 22
 Exercises
Compute the optimum solution for the
following knapsack problem
C=5, n=3. Determine which items are
inserted into the knapsack.
item weight Profit
1 2 $12
2 1 $10
3 3 $20
4 2 $15

21/4/06 RMB/lecture 9 23
Next lecture will be on..

… backtracking algorithm

21/4/06 RMB/lecture 9 24