COL 351: Analysis and Design

of Algorithms

Lecture 9

Keerti Choudhary
Department of CSE, IIT Delhi
 Dynamic Programming
using
Efficient Data-Structures
 Mario Game

Each brick has

exactly one coin

Question: How to collect maximum possible coins if only possible

moves are “Ascending-Right-jumps’’?
 Mario Game

Each brick has

exactly one coin

Question: How to collect maximum possible coins if only possible

moves are “Ascending-Right-jumps’’?
 Mario Game

7
Each brick has
exactly one coin 6

4 4

3 3 3

2 2

Question: How to collect maximum possible coins if only possible

moves are “Ascending-Right-jumps’’?
 Mario Game

An optimal
jump sequence 7
Each brick has
exactly one coin 6

4 4

3 3 3

2 2

Question: How to collect maximum possible coins if only possible

moves are “Ascending-Right-jumps’’?
Longest Increasing Subsequence

Given: An array A = [a1, …, an].

Find: Longest increasing subsequence (LIS) of A.

Longest Increasing Subsequence

Given: An array A = [a1, …, an].

Find: Longest increasing subsequence (LIS) of A.

Eg: A = (8, 5, 7, 3, 9, 4, 11, 2)

Longest Increasing Subsequence

Given: An array A = [a1, …, an].

Find: Longest increasing subsequence (LIS) of A.

Eg: A = (8, 5, 7, 3, 9, 4, 11, 2)

LIS = (5, 7, 9, 11)
Longest Increasing Subsequence

Given: An array A = [a1, …, an].

Find: Longest increasing subsequence (LIS) of A.

Eg: A = (8, 5, 7, 3, 9, 4, 11, 2)

LIS = (5, 7, 9, 11)

Question:
What should be the
sub-problems,
and how to solve them?
Longest Increasing Subsequence

Given: An array A = [a1, …, an].

Find: Longest increasing subsequence (LIS) of A.

Eg: A = (8, 5, 7, 3, 9, 4, 11, 2)

LIS = (5, 7, 9, 11)

Question. Can we have a recurrence to compute the following:

LIS[i] = Length( LIS of A[0,i] )

Ans. No! This will not work for array A = [7, 8, 9, 1, 2, 3, 4, 5, 6].
Longest Increasing Subsequence

Given: An array A = [a1, …, an].

Find: Longest increasing subsequence (LIS) of A.

Eg: A = (8, 5, 7, 3, 9, 4, 11, 2)

LIS = (5, 7, 9, 11)

Question. Can we have a recurrence to compute the following:

LIS[i] = Length( LIS of A[0,i] terminating at ai )

Ans. Yes!
Longest Increasing Subsequence

Given: An array A = [a1, …, an].

Find: Longest increasing subsequence (LIS) of A.

Sub-problem
LIS[i] = Length of LIS of A[0,i] terminating at ai.

Recurrence
LIS[i] = 1 + max {LIS[ j] xj < xi, and j ∈ [1,i − 1]}
Longest Increasing Subsequence

Given: An array A = [a1, …, an].

Find: Longest increasing subsequence (LIS) of A.

Sub-problem
LIS[i] = Length of LIS of A[0,i] terminating at ai.

Recurrence
LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}
Longest Increasing Subsequence

Given: An array A = [a1, …, an].

Find: Longest increasing subsequence (LIS) of A.

Sub-problem
LIS[i] = Length of LIS of A[0,i] terminating at ai.

Recurrence
LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

2 Question:
Time complexity: O(n )
How to improve this?
Data-Structure
Problem
Data Structure Problem

Given: There are n cities (c1, …, cn).

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

1 2 3 4 5 6 7 8
Data Structure Problem

Given: There are n cities (c1, …, cn).

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

1 2 3 4 5 6 7 8

Dynamic: Towers are built in cities in an arbitrary order, with one tower per city.
Data Structure Problem

Given: There are n cities (c1, …, cn).

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

1 2 3 4 5 6 7 8
h = 400

Dynamic: Towers are built in cities in an arbitrary order, with one tower per city.
Data Structure Problem

Given: There are n cities (c1, …, cn).

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

1 2 3 4 5 6 7 8
h = 400 h = 800

Dynamic: Towers are built in cities in an arbitrary order, with one tower per city.
Data Structure Problem

Given: There are n cities (c1, …, cn).

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

1 2 3 4 5 6 7 8
h = 400 h = 800

Dynamic: Towers are built in cities in an arbitrary order, with one tower per city.

Query: What is height of tallest tower in rst i cities?

fi
Data Structure Problem

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

1 2 3 4 5 6 7 8
Data Structure Problem
[1-8]

[1-4] [5-8]

[1-2] [3-4] [5-6] [7-8]

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

1 2 3 4 5 6 7 8
Data Structure Problem
Segment
[1-8]
Tree data-
structure

[1-4] [5-8]

[1-2] [3-4] [5-6] [7-8]

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

1 2 3 4 5 6 7 8

Node [i − j] stores: Height of tallest tower in cities (ci, …, cj)

Data Structure Problem
Segment
Tree data-
structure

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

400 800

Node [i − j] stores: Height of tallest tower in cities (ci, …, cj)

Data Structure Problem
Segment
800 Tree data-
structure
800

400 800

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

400 800

Node [i − j] stores: Height of tallest tower in cities (ci, …, cj)

Data Structure Problem
[1-8]

[1-4] [5-8]

[1-2] [3-4] [5-6] [7-8]

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

600 400 800

Question 1: Time to update nodes of tree (after each tower installation)?

Data Structure Problem
[1-8]

[1-4] [5-8]

[1-2] [3-4] [5-6] [7-8]

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

600 400 800

Question 1: Time to update nodes of tree (after each tower installation)?

Data Structure Problem
[1-8]

[1-4] [5-8]

[1-2] [3-4] [5-6] [7-8]

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

600 400 800

Question 1: Time to update nodes of tree (after each tower installation)?

Ans: O(log n)
Data Structure Problem
[1-8]

[1-4] [5-8]

[1-2] [3-4] [5-6] [7-8]

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

Question 2: Find height of tallest tower in rst i = 6 cities?

fi
Data Structure Problem

[1-4]

[5-6]

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

Question 2: Find height of tallest tower in rst i = 6 cities?

fi
Data Structure Problem

[1-4]

[5-6]

Delhi Mumbai B’lore Jaipur Kolkatta Dubai Chennai NJP

Question 2: Find height of tallest tower in rst i = 6 cities?

Ans: Query O(log2 n) internal nodes in the tree.

fi
 Coming back to
Longest Increasing
subsequence (LIS)
Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i]
Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]
2 3 4 5 7 8 9 11
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i] 1

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i] 1 1

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i] 1 2 1

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i] 1 1 2 1

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i] 1 1 2 1 3

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i] 1 ?? 1 2 1 3

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11

LIS[5] = _____________
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i] 1 ?? 1 2 1 3

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11

LIS[5] = 1+max( rst two leaf nodes) = 2

fi
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i] 1 ?? 1 2 1 3

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11

Remark: Time to nd LIS[ i ] is O(log n).

fi
 Sub-problem

LIS[i] stores length of LIS of A[0,i] terminating at ai.

LIS[i] = 1 + max {LIS[ j] aj < ai, and j ∈ [0, i − 1]}

A = [8, 5, 7, 3, 9, 4, 11, 2]

LIS[i] 1 ?? 1 2 1 3

Sorted A: A[7] A[3] A[5] A[1] A[2] A[0] A[4] A[6]

2 3 4 5 7 8 9 11

Remark: Time to update data-structure is also O(log n).

Exercises

Can you design an e cient algorithm for the Mario Coin collection game, for the
scenario where the bricks contain coins of varying values?

Can you design an O(n log n) time algorithm for Question 4 of Tutorial sheet 5.
ffi