Dynamic Programming

S
 Dynamic Programming

 Dividing a problem into subproblems

 Dynamic programming vs divide and conquer

- Dynamic programming : subproblems are overlapped

- Divide and conquer : subproblems are independent

 Used for finding optimized solutions

 Assembly-Line Scheduling

station 1 station 2 station 3 station 4 station 5 station 6

n=6
Assembly
Line 1 7 9 3 4 8 4

enters exits

Assembly 8 5 6 4 5 7
Line 2
 Assembly-Line Scheduling

Assembly
Line 1 7 9 3 4 8 4

2 3
enters exits

4 2
Assembly 8 5 6 4 5 7
Line 2
 Assembly-Line Scheduling

Assembly
Line 1 7 9 3 4 8 4

2 3
2 3 1 3 4
enters exits
2 1 2 2 1
4 2
Assembly 8 5 6 4 5 7
Line 2
 Assembly-Line Scheduling

Assembly
Line 1 7 9 3 4 8 4
3 40
2 2 3 1 3 4
enters exits
2 1 2 2 1
4 2
41
Assembly 8 5 6 4 5 7
Line 2
 Assembly-Line Scheduling

Assembly
Line 1 7 9 3 4 8 4
3 39
2 2 3 1 3 4
enters exits
2 1 2 2 1
4 2
Assembly 8 5 6 4 5 7
Line 2
 Assembly-Line Scheduling

station 1 station 2 station 3 station 4 station 5 station 6 number

of cases
Assembly
Line 1 7 9 3 4 8 4 = 2n

enters exits

Assembly 8 5 6 4 5 7
Line 2
 Assembly-Line Scheduling

T1,1 T1,2 T1,3 T1,4 T1,5 T1,6

Assembly
Line 1

T
enters exits

Assembly
Line 2
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
Assembly-Line Scheduling

T1,5 T1,6
8 4 T = min( T1,6 + 3, T2,6 + 2)

3 Divide and Conquer?

4 T
exits T1,6 = min( T1,5 + 4,
T2,5 + 1 + 4)
1
2
T2,6 = min( T1,5 + 4 + 7,
5 7 T2,5 + 7)

T2,5 T2,6
Assembly-Line Scheduling

T1,5 T1,6
8 4 T = min( T1,6 + 3, T2,6 + 2)

3 Divide and Conquer?

4 T
exits T1,6 = min( T1,5 + 4,
T2,5 + 1 + 4)
1
2
T2,6 = min( T1,5 + 4 + 7,
5 7 T2,5 + 7)

T2,5 T2,6
 Assembly-Line Scheduling

9
T1,1 T1,2 T1,3
Assembly
Line 1 7 9 3 T1,1 = 9
T2,1 = 12
2 2 3
enters
2 1
4
Assembly 8 5 6
Line 2
T2,1 T2,2 T2,3
12
 Assembly-Line Scheduling

9 18
T1,1 T1,2 T1,3
Assembly
Line 1 7 9 3 T1,1 = 9
T2,1 = 12
2 2 3 T1,2 = min(T1,1 + 9,
enters T2,1 + 2 + 9)
2 1
4
Assembly 8 5 6
Line 2
T2,1 T2,2 T2,3
12
 Assembly-Line Scheduling

9 18
T1,1 T1,2 T1,3
Assembly
Line 1 7 9 3 T1,1 = 9
T2,1 = 12
2 2 3 T1,2 = 18
enters T2,2 = min(T1,1 + 2 + 5,
2 1 T2,1 + 5)
4
Assembly 8 5 6
Line 2
T2,1 T2,2 T2,3
12 16
 Assembly-Line Scheduling

9 18 20 24 32 35
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6
Assembly
Line 1 7 9 3 4 8 4
3 38
2 2 3 1 3 4 T
enters exits
2 1 2 2 1
4 2
Assembly 8 5 6 4 5 7
Line 2
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
12 16 22 25 30 37
Assembly-Line Scheduling

32 35
T1,5 T1,6
8 4 T = T1,6 + 3 ? or T2,6 + 2 ?

3 38
T T1,6 + 3 = T
4
exits T2,6 + 2 > T
1
2
5 7
T2,5 T2,6
30 37
Assembly-Line Scheduling

32 35
T1,5 T1,6
8 4 T = T1,6 + 3 ? or T2,6 + 2 ?

3 38
T T1,6 + 3 = T
4
exits T2,6 + 2 > T
1
2
5 7
T2,5 T2,6
30 37
Assembly-Line Scheduling

32 35
T1,5 T1,6
8 4 T1,6 = T1,5 + 4 ?
or T2,5 + 1 + 4 ?
3 38
4 T
T1,5 + 4 > T1,6
exits
T2,5 + 1 + 4 = T1,6
1
2
5 7
T2,5 T2,6
30 37
Assembly-Line Scheduling

32 35
T1,5 T1,6
8 4 T1,6 = T1,5 + 4 ?
or T2,5 + 1 + 4 ?
3 38
4 T
T1,5 + 4 > T1,6
exits
T2,5 + 1 + 4 = T1,6
1
2
5 7
T2,5 T2,6
30 37
 Assembly-Line Scheduling

9 18 20 24 32 35
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6
Assembly
Line 1 7 9 3 4 8 4
3 38
2 2 3 1 3 4 T
enters exits
2 1 2 2 1
4 2
Assembly 8 5 6 4 5 7
Line 2
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
12 16 22 25 30 37
Assembly-Line Scheduling

T1,5 T1,6 1. Analyze the problem

T
exits T = min( T1,6 + 3, T2,6 + 2)

T2,5 T2,6
Assembly-Line Scheduling

T1,5 T1,6 1. Analyze the problem

T 2. Find a recursive solution.

exits T = min( T1,6 + 3, T2,6 + 2)

T1,6 = min( T1,5 + 4,

T2,5 T2,6 T2,5 + 1 + 4)
T2,6 = min( T1,5 + 4 + 7,
T2,5 + 7)
 Assembly-Line Scheduling

9 18 20 24 32 35
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6 1. Analyze the problem.
2. Find a recursive solution.
T 3. Compute the fastest time.
38

T2,1 T2,2 T2,3 T2,4 T2,5 T2,6

12 16 22 25 30 37
 Assembly-Line Scheduling

9 18 20 24 32 35
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6 1. Analyze the problem.
2. Find a recursive solution.
T 3. Compute the fastest time.
38 4. Construct the fastest way.

T2,1 T2,2 T2,3 T2,4 T2,5 T2,6

12 16 22 25 30 37
 Assembly-Line Scheduling

9 18 20 24 32 35
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6 1. Analyze the problem.
2. Find a recursive solution.
T 3. Compute the fastest time.
38 4. Construct the fastest way.

T2,1 T2,2 T2,3 T2,4 T2,5 T2,6 This algorithm takes time.
12 16 22 25 30 37 Brute force :
 When can we use
Dynamic Programming?

 Optimal substructures

 Overlapping subproblems
Overlapping Subprograms

T1,5 T1,6
T
8 4
3 T1,6 T2,6
4 T
exits
1 T1,5 T2,5 T1,5 T2,5
2
5 7 T1,4 T2,4 T1,4 T2,4 T1,4 T2,4 T1,4 T2,4
T2,5 T2,6
Overlapping Subprograms

T1,5 T1,6
T
8 4
3 T1,6 T2,6
4 T
exits
1 T1,5 T2,5 T1,5 T2,5
2
5 7 T1,4 T2,4 T1,4 T2,4 T1,4 T2,4 T1,4 T2,4
T2,5 T2,6
 Optimal Substructures

 Problems should be divided into subproblems.

 Optimal solution : from optimal subproblem solutions

One part of the optimal solution to the problem

should be an optimal solution to the subproblem

solutions to the subproblems should be independent

 Optimal Substructures

1. Shortest Path Problem 2. Longest Simple Path Problem

d d

s s
 Optimal Substructures

1. Shortest Path Problem 2. Longest Simple Path Problem

d d

s s
w
w
 Memoization

 Ordinary

- Bottom-up strategy

 Memoization

- Top-down strategy
 Memoization

Memoization Bottom Up
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6 T1,1 T1,2 T1,3 T1,4 T1,5 T1,6
- - - - - -
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6 T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
- - - - - -
 Memoization

T1,6 T2,6

T1,5 T2,5
Memoization
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6 T1,4 T2,4
- - - - - -
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
… …
- - - - - -
 Memoization

T1,3

Memoization T1,2 T2,2

T1,1 T1,2 T1,3 T1,4 T1,5 T1,6

9 - - - - -
T1,1 T2,1
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
12 - - - - -
 Memoization

T1,3

Memoization T1,2 T2,2

T1,1 T1,2 T1,3 T1,4 T1,5 T1,6

9 18 - - - -
T1,1 T2,1
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
12 - - - - -
 Memoization

T1,3

Memoization T1,2 T2,2

T1,1 T1,2 T1,3 T1,4 T1,5 T1,6

9 18 - - - -
T1,1 T2,1 T1,1 T2,1
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
12 16 - - - -
 Memoization

T1,4

Memoization T1,3 T2,3

T1,1 T1,2 T1,3 T1,4 T1,5 T1,6

9 18 20 - - -
T1,2 T2,2 T1,2 T2,2
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
12 16 - - - - T1,1 T2,1 T1,1 T2,1
 Memoization

T1,6 T2,6

T1,5 T2,5
Memoization
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6 T1,4 T2,4 T1,4 T2,4
9 18 20 24 - -
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
… …
12 16 22 - - - T1,3 T2,3
 Memoization

T1,6 T2,6

T1,5 T2,5
Memoization
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6 T1,4 T2,4 T1,4 T2,4
9 18 20 24 32 -
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
… … … …
12 16 22 25 - -
 Memoization

T1,6 T2,6

T1,5 T2,5 T1,5 T2,5

Memoization
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6 T1,4 T2,4 T1,4 T2,4
9 18 20 24 32 35
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
… … … …
12 16 22 25 30 -
 Memoization

T = 38

T1,6 T2,6

T1,5 T2,5 T1,5 T2,5

Memoization
T1,1 T1,2 T1,3 T1,4 T1,5 T1,6 T1,4 T2,4 T1,4 T2,4
9 18 20 24 32 35
T2,1 T2,2 T2,3 T2,4 T2,5 T2,6
… … … …
12 16 22 25 30 37
 Why Memoization?

 Ordinary Dynamic Programming (Bottom-Up strategy)

- when all subproblems need to be solved

 Memoization

- when some subproblems do not need to be solved

Longest Common Subsequence

 DNA sequences : composed of four components

– {A, C, G, T}

S1 = ACCGGTCGTGCGCGGAAGCCGGCCGAA
S2 = GTCGTTCGGAATGCCGTTGCTCTGTAAA
 How similar are they?
Longest Common Subsequence

S1 = ACCGGTCGTGCGCGGAAGCCGGCCGAA

S2 = GTCGTTCGGAATGCCGTTGCTCTGTAAA

Longest Common Sequence :

GTCGTCGGAAGCCGGCCGAA
Longest Common Subsequence

X = <x1, x2, …, xm>, Y = <y1, y2, …, yn>

Xi = <x1, x2, …, xi> for i = 1 to m

Yj = <y1, y2, …, yj> for j = 1 to n

Zi,j = <z1, z2, …, zk> for i = 1 to m, j = 1 to n

Longest common subsequence of Xi and Yj
Longest Common Subsequence

1. Analyze the problem.

X = <x1, x2, …, xm>, Y = <y1, y2, …, yn>

Xi = <x1, x2, …, xi> Yj = <y1, y2, …, yj>
Zi,j = <z1, z2, …, zk> (Longest common subsequence)
Zi-1,j = <z1, z2, …, zk-1> when xi used as zk
or
<z1, z2, …, zk> otherwise
Zi,j-1 = <z1, z2, …, zk> or <z1, z2, …, zk-1>
Longest Common Subsequence

1. Analyze the problem.

X = <x1, x2, …, xm>, Y = <y1, y2, …, yn>

Xi = <x1, x2, …, xi> Yj = <y1, y2, …, yj>
Zi,j = <z1, z2, …, zk> (Longest common subsequence)
If xi =/ yj then zk =/ xi or zk = /yj
Zi-1,j = <z1, z2, …, zk-1> when xi used as zk
or
<z1, z2, …, zk> otherwise
Longest Common Subsequence

1. Analyze the problem.

X = <x1, x2, …, xm>, Y = <y1, y2, …, yn>

Xi = <x1, x2, …, xi> Yj = <y1, y2, …, yj>
Zi,j = <z1, z2, …, zk> (Longest common subsequence)
If xi =/ yj then zk =/ xi or zk = /yj
Zi,j = Zi-1,j or Zi,j-1
Longest Common Subsequence

1. Analyze the problem.

X = <x1, x2, …, xm>, Y = <y1, y2, …, yn>

Xi = <x1, x2, …, xi> Yj = <y1, y2, …, yj>
Zi,j = <z1, z2, …, zk> (Longest common subsequence)
If xi =/ yj then Zi,j = Zi-1,j or Zi,j-1 Optimal
If xi = yj then
If xi = yj is not used as zk then Zi,j = Zi-1,j or Zi,j-1
If xi = yj is used as zk then Zi,j = Zi-1,j-1 + zk
Longest Common Subsequence

1. Analyze the problem.

X = <x1, x2, …, xm>, Y = <y1, y2, …, yn>

Xi = <x1, x2, …, xi> Yj = <y1, y2, …, yj>
Zi,j = <z1, z2, …, zk> (Longest common subsequence)
If xi =/ yj then Zi,j = Zi-1,j or Zi,j-1 ci,j = max(ci-1,j, ci,j-1)
If xi = yj then Zi,j = Zi-1,j-1 + zk ci,j = ci-1,j-1 + 1
Longest Common Subsequence

2. Find a recursive solution.

If i = 0 or j = 0 then ci,j = 0
Else
If xi =/ yj then ci,j = max(ci-1,j, ci,j-1)
If xi = yj then ci,j = ci-1,j-1 + 1
Longest Common Subsequence

3. Compute the fastest time.

C1,1 C1,2 . . . C1,n If xi =/ yj then

C2,1 C2,2 . . . C2,n ci,j = max(ci-1,j, ci,j-1)
. . . . If xi = yj then
. . . .
ci,j = ci-1,j-1 + 1
. . . .

Cm,1 Cm,2 . . . Cm,n

Longest Common Subsequence

4. Construct the common

subsequence.
C1,1 C1,2 . . . C1,n
C2,1 C2,2 . . . C2,n
. . . .
. . . .
. . . .

Cm,1 Cm,2 . . . Cm,n

Longest Common Subsequence

4. Construct the common

subsequence.
Ci-2,j-2 Ci-2,j-1 Ci-2,j
If xi = yj then
Ci-1,j-2 Ci-1,j-1 Ci-1,j move to ci-1,j-1
Else
Ci,j-2 Ci,j-1 Ci,j
If ci,j = ci-1,j then
move to ci-1,j
If xi =/ yj then If ci,j = ci,j-1 then
ci,j = max(ci-1,j, ci,j-1) move to ci,j-1
If xi = yj then one element of
ci,j = ci-1,j-1 + 1 The longest common subsequence
 Dynamic Programming

 Dividing a problem into subproblems.

 For optimization problems.

 Optimal substructures / overlapping subproblems

 Process of dynamic programmings

1. Analyze the problem.
2. Find a recursive solution.
3. Compute the fastest time/cost.
4. Construct the fastest path.