Dynamic Programming

Part II

Md. Bakhtiar Hasan

Lecturer
Department of Computer Science and Engineering
Islamic University of Technology
Designing A General Solution

Define Subproblems

MBH (CSE, IUT) Dynamic Programming 2 / 10

Designing A General Solution

Define Subproblems
Relate Subproblems

MBH (CSE, IUT) Dynamic Programming 2 / 10

Designing A General Solution

Define Subproblems
Relate Subproblems
Identify Base Cases

MBH (CSE, IUT) Dynamic Programming 2 / 10

Designing A General Solution

Define Subproblems
Relate Subproblems
Identify Base Cases
Compute Solution from Subproblems

MBH (CSE, IUT) Dynamic Programming 2 / 10

Designing A General Solution

Define Subproblems
Relate Subproblems
Identify Base Cases
Compute Solution from Subproblems
Analyze Running Time

MBH (CSE, IUT) Dynamic Programming 2 / 10

Define Subproblems

Describe the meaning of subproblems

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters
Subset of input

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters
Subset of input
• Prefixes/Suffixes

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters
Subset of input
• Prefixes/Suffixes
• Contiguous subsequences

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters
Subset of input
• Prefixes/Suffixes
• Contiguous subsequences
Might record partial state

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters
Subset of input
• Prefixes/Suffixes
• Contiguous subsequences
Might record partial state
• Add subproblems by incrementing auxiliary variables

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters
Subset of input
• Prefixes/Suffixes
• Contiguous subsequences
Might record partial state
• Add subproblems by incrementing auxiliary variables

Examples
Fibonacci Series: F (i), the ith Fibonacci Number

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters
Subset of input
• Prefixes/Suffixes
• Contiguous subsequences
Might record partial state
• Add subproblems by incrementing auxiliary variables

Examples
Fibonacci Series: F (i), the ith Fibonacci Number
Shortest Path

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters
Subset of input
• Prefixes/Suffixes
• Contiguous subsequences
Might record partial state
• Add subproblems by incrementing auxiliary variables

Examples
Fibonacci Series: F (i), the ith Fibonacci Number
Shortest Path
1 d(v ), the shortest path weight from s to v

MBH (CSE, IUT) Dynamic Programming 3 / 10

Define Subproblems

Describe the meaning of subproblems

• Describe the parameters
Subset of input
• Prefixes/Suffixes
• Contiguous subsequences
Might record partial state
• Add subproblems by incrementing auxiliary variables

Examples
Fibonacci Series: F (i), the ith Fibonacci Number
Shortest Path
1 d(v ), the shortest path weight from s to v
2 d(v , k), the shortest path weight from s to v using at most k edges

MBH (CSE, IUT) Dynamic Programming 3 / 10

Relate Subproblems

State the topological order

MBH (CSE, IUT) Dynamic Programming 4 / 10

Relate Subproblems

State the topological order→ Guess if required

MBH (CSE, IUT) Dynamic Programming 4 / 10

Relate Subproblems

State the topological order→ Guess if required

Argue the relations are acyclic

MBH (CSE, IUT) Dynamic Programming 4 / 10

Relate Subproblems

State the topological order→ Guess if required

Argue the relations are acyclic
Form a DAG

MBH (CSE, IUT) Dynamic Programming 4 / 10

Relate Subproblems

State the topological order→ Guess if required

Argue the relations are acyclic
Form a DAG

Examples
Fibonacci Series: F (i) = F (i − 1) + F (i − 2)

MBH (CSE, IUT) Dynamic Programming 4 / 10

Relate Subproblems

State the topological order→ Guess if required

Argue the relations are acyclic
Form a DAG

Examples
Fibonacci Series: F (i) = F (i − 1) + F (i − 2)
Shortest Paths

MBH (CSE, IUT) Dynamic Programming 4 / 10

Relate Subproblems

State the topological order→ Guess if required

Argue the relations are acyclic
Form a DAG

Examples
Fibonacci Series: F (i) = F (i − 1) + F (i − 2)
Shortest Paths
1 d(v ) = min{d(u) + cost[u][v ] | (u, v ) ∈ E }

MBH (CSE, IUT) Dynamic Programming 4 / 10

Relate Subproblems

State the topological order→ Guess if required

Argue the relations are acyclic
Form a DAG

Examples
Fibonacci Series: F (i) = F (i − 1) + F (i − 2)
Shortest Paths
1 d(v ) = min{d(u) + cost[u][v ] | (u, v ) ∈ E }(Assume graph is acyclic)

MBH (CSE, IUT) Dynamic Programming 4 / 10

Relate Subproblems

State the topological order→ Guess if required

Argue the relations are acyclic
Form a DAG

Examples
Fibonacci Series: F (i) = F (i − 1) + F (i − 2)
Shortest Paths
1 d(v ) = min{d(u) + cost[u][v ] | (u, v ) ∈ E }(Assume graph is acyclic)
2 d(v , k) = min{d(u, k − 1) + cost[u][v ] | (u, v ) ∈ E }

MBH (CSE, IUT) Dynamic Programming 4 / 10

Identify Base Cases

Solutions for independent subproblems

MBH (CSE, IUT) Dynamic Programming 5 / 10

Identify Base Cases

Solutions for independent subproblems

Examples
Fibonacci Series: F (1) = F (2) = 1

MBH (CSE, IUT) Dynamic Programming 5 / 10

Identify Base Cases

Solutions for independent subproblems

Examples
Fibonacci Series: F (1) = F (2) = 1
Shortest Paths:

MBH (CSE, IUT) Dynamic Programming 5 / 10

Identify Base Cases

Solutions for independent subproblems

Examples
Fibonacci Series: F (1) = F (2) = 1
Shortest Paths:
• d(s) = δ(s, s) = 0 and d(v ) = ∞ for all v 6= s without incoming edge

MBH (CSE, IUT) Dynamic Programming 5 / 10

Identify Base Cases

Solutions for independent subproblems

Examples
Fibonacci Series: F (1) = F (2) = 1
Shortest Paths:
• d(s) = δ(s, s) = 0 and d(v ) = ∞ for all v 6= s without incoming edge
• d(s, 0) = 0 and d(v , 0) = ∞ for all v 6= s without incoming edge

MBH (CSE, IUT) Dynamic Programming 5 / 10

Compute Solution from Subproblems

Use either top-down/bottom-up approach

MBH (CSE, IUT) Dynamic Programming 6 / 10

Compute Solution from Subproblems

Use either top-down/bottom-up approach

Can use parent pointers to store the states

MBH (CSE, IUT) Dynamic Programming 6 / 10

Compute Solution from Subproblems

Use either top-down/bottom-up approach

Can use parent pointers to store the states

Examples
Fibonacci Series: Find F (n)

MBH (CSE, IUT) Dynamic Programming 6 / 10

Compute Solution from Subproblems

Use either top-down/bottom-up approach

Can use parent pointers to store the states

Examples
Fibonacci Series: Find F (n)
Shortest Paths:

MBH (CSE, IUT) Dynamic Programming 6 / 10

Compute Solution from Subproblems

Use either top-down/bottom-up approach

Can use parent pointers to store the states

Examples
Fibonacci Series: Find F (n)
Shortest Paths:
1 Find d(v ) for each v ∈ V

MBH (CSE, IUT) Dynamic Programming 6 / 10

Compute Solution from Subproblems

Use either top-down/bottom-up approach

Can use parent pointers to store the states

Examples
Fibonacci Series: Find F (n)
Shortest Paths:
1 Find d(v ) for each v ∈ V
2 Find δk (s, v ) for k = |V | − 1 and v ∈ V

MBH (CSE, IUT) Dynamic Programming 6 / 10

Analyze Running Time

Work done in each subproblem

MBH (CSE, IUT) Dynamic Programming 7 / 10

Analyze Running Time

Work done in each subproblem

• is not equal:
P
work(x)
x∈X

MBH (CSE, IUT) Dynamic Programming 7 / 10

Analyze Running Time

Work done in each subproblem

• is not equal:
P
work(x)
x∈X
• is equal: |X | × W

MBH (CSE, IUT) Dynamic Programming 7 / 10

Analyze Running Time

Work done in each subproblem

• is not equal:
P
work(x)
x∈X
• is equal: |X | × W

Examples
Fibonacci Series: O(n)

MBH (CSE, IUT) Dynamic Programming 7 / 10

Analyze Running Time

Work done in each subproblem

• is not equal:
P
work(x)
x∈X
• is equal: |X | × W

Examples
Fibonacci Series: O(n)
Shortest Paths:

MBH (CSE, IUT) Dynamic Programming 7 / 10

Analyze Running Time

Work done in each subproblem

• is not equal:
P
work(x)
x∈X
• is equal: |X | × W

Examples
Fibonacci Series: O(n)
Shortest Paths:
1 O(V + E )

MBH (CSE, IUT) Dynamic Programming 7 / 10

Analyze Running Time

Work done in each subproblem

• is not equal:
P
work(x)
x∈X
• is equal: |X | × W

Examples
Fibonacci Series: O(n)
Shortest Paths:
1 O(V + E )
2 O(VE )

MBH (CSE, IUT) Dynamic Programming 7 / 10

Problem At Hand

Given a rod of length n

MBH (CSE, IUT) Dynamic Programming 8 / 10

Problem At Hand

Given a rod of length n

Price p(i) for rod piece of length i (1 ≤ i ≤ n)

MBH (CSE, IUT) Dynamic Programming 8 / 10

Problem At Hand

Given a rod of length n

Price p(i) for rod piece of length i (1 ≤ i ≤ n)
Cut rod to maximize the profit

MBH (CSE, IUT) Dynamic Programming 8 / 10

Problem At Hand

Given a rod of length n

Price p(i) for rod piece of length i (1 ≤ i ≤ n)
Cut rod to maximize the profit

1 5 8 9
1 2 3 4

MBH (CSE, IUT) Dynamic Programming 8 / 10

Problem At Hand

Given a rod of length n

Price p(i) for rod piece of length i (1 ≤ i ≤ n)
Cut rod to maximize the profit

1 5 8 9
1 2 3 4

MBH (CSE, IUT) Dynamic Programming 8 / 10

Solution Idea

Subproblem

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution
• Find x(n)

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution
• Find x(n)
• Store choices to reconstruct

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution
• Find x(n)
• Store choices to reconstruct
I Rod length i, Optimal choice j?

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution
• Find x(n)
• Store choices to reconstruct
I Rod length i, Optimal choice j?→ (i − j) remains

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution
• Find x(n)
• Store choices to reconstruct
I Rod length i, Optimal choice j?→ (i − j) remains
Time

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution
• Find x(n)
• Store choices to reconstruct
I Rod length i, Optimal choice j?→ (i − j) remains
Time
• # of subproblems: n

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution
• Find x(n)
• Store choices to reconstruct
I Rod length i, Optimal choice j?→ (i − j) remains
Time
• # of subproblems: n
• Time/subproblem: O(i)

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution
• Find x(n)
• Store choices to reconstruct
I Rod length i, Optimal choice j?→ (i − j) remains
Time
• # of subproblems: n
• Time/subproblem: O(i)→ Not same for all subproblems

MBH (CSE, IUT) Dynamic Programming 9 / 10

Solution Idea

Subproblem
• x(i) Maximum value obtained by cutting rod of length i
Relate
• Left-most cut has some length → Guess
• x(i) = max{p(j) + x(i − j) | j ∈ {1, . . . , i}}
• x(i) only depend on smaller i, acyclic
Base
• Length zero rod has no value!
• x(0) = 0
Solution
• Find x(n)
• Store choices to reconstruct
I Rod length i, Optimal choice j?→ (i − j) remains
Time
• # of subproblems: n
• Time/subproblem: O(i)→ Not same for all subproblems
• Sum up: 1 + 2 + · · · + n = O(n2 )
MBH (CSE, IUT) Dynamic Programming 9 / 10
Reading

CLRS 15.1

MBH (CSE, IUT) Dynamic Programming 10 / 10