CS213/293 Data Structure and Algorithms 2023

Lecture 10: Pattern matching

Instructor: Ashutosh Gupta

IITB India

Compile date: 2023-09-03

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 1
 Topic 10.1

Pattern matching problem

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 2
Pattern matching

Definition 10.1
In a pattern-matching problem, we need to find the position of all occurrences of a pattern string
P in a string T .

Usage:
▶ Text editor
▶ DNA sequencing

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 3
Example : Näive approach for pattern matching
Example 10.1
Consider the following text T and pattern P. We try to match the pattern in every position.

T x y z x y x x y x y p x

P x y x y

x y x y

x y x y

x y x y

x y x y

x y x y

x y x y
Running time complexity is O(|T ||P|).
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 4
Wasteful attempts of matching.
i
T x y z x y x x y x y p x

P x y x y

x y x y

x y x y

Should we have tried to match the second and third positions?

No.

Let us suppose we failed to match position i of T and position 2 of P.

▶ We know that T [i − 1] = y . Therefore, there is no match starting at i − 1. (Why?)

▶ We know that T [i] ̸= x. Therefore, there is no match starting at i. (Why?)

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 5
Shifting the pattern
Let us suppose at position i of T and j of P the matching fails.

T ... ...

Let us suppose we want to resume the search by only updating j.

If we assign j some value k, we are shifting the pattern forward by j − k.

Exercise 10.1
What is the meaning of k = j − 1, k = 0, or k = −1?
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 6
What is a good value of k?
We know T [i − j : i − 1] = P[0 : j − 1] and T [i] ̸= P[j].

T ... ...

P
P[j − k : j − 1] j

P[0 : k − 1]
k

We must have P[0 : k − 1] = P[j − k : j − 1] and P[j] ̸= P[k](Why?).

Exercise 10.2
Should we choose the largest k or smallest k?
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 7
The largest k implies the minimum shift

We choose the largest k such that

P[0 : k − 1] = P[j − k : j − 1] and P[j] ̸= P[k].

k only depends on P and j.

Since P is typically small, we may pre-compute array h such that h[j] = k.

Example 10.2
P x y x y P x y x z

h −1 0 −1 0 2 h −1 0 −1 1 0

We can compute h in O(|P|) time. We will discuss this later.

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 8
 Knuth–Morris–Pratt algorithm
Algorithm 10.1: KMP(string T,string P)
Running time complexity:
1 i := 0; j := 0; found := ∅;
▶ if branch will only execute ≤ |T | times.
2 h := KMPTable(P);
3 while i < |T | do
▶ How do we bound the number of
4 if P[j] = T [i] then
iterations where the else branch does not
5 i := i + 1; j := j + 1; increment i?
6 if j = |P| then 1. The else branch reduces j.
7 found.insert(i − j); 2. Since j ≥ 0 at loop head, the no. of
8 j = h[j]; reductions of j ≤ no. of the increments
of j.
9 else
3. i and j are incremented together.
10 j = h[j]; 4. no. of reductions of j ≤ no. of
11 if j < 0 then increments of i.
12 i := i + 1; j := j + 1; 5. no. of reductions of j ≤ |T |
▶ O(|T |) algorithm
13 return found
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 9
 Topic 10.2

How to compute array h?

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 10
Recall: the definition of h

For a pattern P, h[j] is the largest k such that

P[0 : k − 1] = P[j − k : j − 1] and P[j] ̸= P[k].

We use KMP like algorithm again to compute h.

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 11
Self-matching: We use KMP again for computing h

For largest j such that P[i − j : i − 1] = P[0 : j − 1] and P[i] ̸= P[j].

P
j

We assign h[i] = j.

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 12
Self-matching: Moving the pattern forward
After the mismatch, we need to move the pattern forward as little as possible.

P
j

We must have computed h for earlier indexes. Therefore, j := h[j].

Exercise 10.3
Why the value of h[j] be available?
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 13
Self-matching: if no disagreement
Let us consider the case when matching continues. How should we assign h[i]?

P
j

h[i] := j may not be efficient.

If the suffix of part of T does not match with P[0 : i] then it will also not match with P[0 : j].

We will be jumping again to h[j]. We should directly assign h[i] = h[j].

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 14
 Computing h array

Algorithm 10.2: KMPTable(string P)

i := 1; j := 0; h[0] := − 1;
while i < |P| do
if P[j] ̸= P[i] then
4 h[i] := j;
5 while j ≥ 0 and P[j] ̸= P[i] do
6 j := h[j]; // Moving forward the pattern in minimum steps as in KMP
else
8 h[i] := h[j];
i := i + 1; j := j + 1;
h[|P|] := j;
return h

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 15
 Topic 10.3

Problem

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 16
Exercise: compute h

Exercise 10.4
Compute array h for pattern ”babbaabba”.

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 17
 Exercise: version of KMPtable

Exercise 10.5
Is the following version of KMPtable correct?

Algorithm 10.3: KMPTableV2(string P)

i := 1; j := 0; h[0] := − 1;
while i < |P| do
h[i] := j;
while j ≥ 0 and P[j] ̸= P[i] do
5 j := h[j]; // Moving forward the pattern in minimum steps as in KMP
i := i + 1; j := j + 1;
h[|P|] := j;
return h

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 18
Exercise: compute h(i)

Exercise 10.6
Suppose that there is a letter z in P of length n such that it occurs in only one place, say k, which
is given in advance. Can we optimize the computation of h?

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 19
 End of Lecture 10

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 20