String Matching

Naive String Matching

Rabin-Karp Matcher
String Matching with Finite Automata
Knuth-Morris-Pratt Algorithm

Chapter 32: Slide – 1

 String Matching

⊲ string matching Input is a pattern P [1..m] and text T [1..n].

rabin karp
rabin karp 2
rabin karp 3
Find a shift s s.t. P [i] = T [s + i] for 1 ≤ i ≤ m.
rabin karp 4
fsa 1
Applications: document search, grep, DNA
fsa 2
fsa 3
fsa 4
fsa 5
Naive-String-Matcher(T, P )
fsa 6
fsa 7
n ← T.length, m ← P.length
kmp 1
kmp 2
for s ← 0 to n − m
kmp 3
kmp 4
if P [1..m] = T [s + 1..s + m]
kmp 5
pattern occurs with shift s

Naive-String-Matcher is O(nm).
Consider P = aaaaa and T = aaaabaaaab. . .

Chapter 32: Slide – 2

 Rabin-Karp String Matching
string matching P [1..m] can be converted to a number:
⊲ rabin karp
rabin karp 2 m−1
rabin karp 3
X
rabin karp 4
fsa 1
p= P [m − i] ∗ di
fsa 2 i=0
fsa 3
fsa 4
fsa 5
where each character is in radix-d notation.
fsa 6
fsa 7
Similarly, T [s + 1 . . . s + m] can be converted:
kmp 1
kmp 2 m−1
kmp 3
X
kmp 4 t= T [s + m − i] ∗ di
kmp 5
i=0

Idea: test p mod q = t mod q before the more

expensive P [1 . . . m] = T [s + 1 . . . s + m].
Use a prime number for q.

Chapter 32: Slide – 3

 Rabin-Karp Example
string matching
rabin karp
⊲ rabin karp 2
rabin karp 3
rabin karp 4
fsa 1
fsa 2
fsa 3
fsa 4
fsa 5
fsa 6
fsa 7
kmp 1
kmp 2
kmp 3
kmp 4
kmp 5

Chapter 32: Slide – 4

 Rabin-Karp Matcher
string matching
rabin karp
rabin karp 2
Rabin-Karp-Matcher(T, P, d, q)
⊲ rabin karp 3 n ← T.length, m ← P.length
rabin karp 4
fsa 1
fsa 2
p ← 0, t ← 0, h ← dm mod q
fsa 3
fsa 4
for i ← 1 to m
fsa 5
fsa 6
p ← (d ∗ p + P [i]) mod q
fsa 7 t ← (d ∗ t + T [i]) mod q
kmp 1
kmp 2 for s ← 0 to n − m
kmp 3
kmp 4 if s > 0
kmp 5
t ← (d∗t + T [s+m] − T [s]∗h) mod q
if p = t and P [1..m] = T [s + 1..s + m]
pattern occurs with shift s

Chapter 32: Slide – 5

 Average Case Analysis of Rabin-Karp
string matching
rabin karp Probability Model:
rabin karp 2
rabin karp 3
Suppose there are v valid shifts.
⊲ rabin karp 4 If s is not a valid shift, then p mod q = t mod q
fsa 1
fsa 2
fsa 3
with probability 1/q.
fsa 4
fsa 5
fsa 6 O(m) per valid shift. O(m) per “spurious hit.”
fsa 7
kmp 1 Expected number of spurious hits is O(n/q).
kmp 2
kmp 3 O(1) per iteration otherwise.
kmp 4
kmp 5

Expected running time is O(m(v + n/q) + n),

which is O(mv + n) if q ≥ m.

Chapter 32: Slide – 6

 Finite State Automata
string matching A finite state automaton is defined by:
rabin karp
rabin karp 2 Q, a set of states
rabin karp 3
rabin karp 4 q0 ∈ Q, the start state
⊲ fsa 1 A ⊆ Q, the accepting states
fsa 2
fsa 3
fsa 4
Σ, the input alphabet
fsa 5
fsa 6
δ, the transition function, from Q × Σ to Q
fsa 7
kmp 1
kmp 2
kmp 3
kmp 4
kmp 5

Chapter 32: Slide – 7

 FSA Idea for String Matching
string matching Start in state qo .
rabin karp
rabin karp 2 Perform a transition from q0 to q1 if next character
rabin karp 3
rabin karp 4 of T = P [1].
fsa 1
⊲ fsa 2
fsa 3
fsa 4
State qi means first i characters of P match.
fsa 5
fsa 6
Transition from qi to qi+1 if the next character of
fsa 7
kmp 1
T = P [i + 1].
kmp 2
kmp 3
kmp 4 Transition Function for P = aba
kmp 5

State
0 1 2 3 0 a 1 b 2 a 3
Inputs a 1 ? 3 ?
b ? 2 ? ?

Chapter 32: Slide – 8

 FSA Matcher (Incomplete)
string matching
rabin karp
rabin karp 2
FSA-Matcher(T, P )
rabin karp 3
rabin karp 4
n ← T.length, m ← P.length
fsa 1
fsa 2
q ← 0 // q is the state of the FSA.
⊲ fsa 3
for s ← 1 to n
fsa 4
fsa 5
fsa 6
if q < m and T [s] = P [q + 1]
fsa 7
kmp 1
q ← q+1
kmp 2
kmp 3
else q ← ???
kmp 4
kmp 5
if q = m
pattern occurs with shift s − m

Cannot simply reset state to 0.

Consider P = ab and T = aab

Chapter 32: Slide – 9

 Delta Transition for String Matching

string matching  qi+1 if i < m and T [s] = P [i + 1]
rabin karp 

rabin karp 2 
if j is the maximum value
rabin karp 3


rabin karp 4

fsa 1 δ(qi , T [s]) = qj+1 such that T [s] = P [j + 1] and
fsa 2
T [s − j..s − 1] = P [1..j]
fsa 3

⊲


fsa 4 

fsa 5 
q0 otherwise
fsa 6

fsa 7
kmp 1
kmp 2
kmp 3

kmp 4 
 qi+1 if i < m and x = P [i + 1]
kmp 5 

if j is the maximum value



δ(qi , x) = qj+1 such that x = P [j + 1] and
P [i − j + 1..i] = P [1..j]






q0 otherwise


Chapter 32: Slide – 10

 FSA for Matching bbbba
string matching
rabin karp
rabin karp 2
a a a
rabin karp 3
rabin karp 4 a b
fsa 1
fsa 2 a 0 1 2 3 4 a 5
fsa 3
fsa 4
b b b b
⊲ fsa 5
b
fsa 6
fsa 7
kmp 1
kmp 2
kmp 3
Transition Function
kmp 4
kmp 5
States
0 1 2 3 4 5
Inputs a 0 0 0 0 5 0
b 1 2 3 4 4 1

Chapter 32: Slide – 11

 FSA for Matching ababa
string matching
rabin karp
rabin karp 2 b b b
rabin karp 3
rabin karp 4
fsa 1
b 0 a 1 b 2 a 3 b 4 a 5
fsa 2
fsa 3
a a
fsa 4
fsa 5
⊲ fsa 6
a
fsa 7
kmp 1
kmp 2 Transition Function
kmp 3
kmp 4
kmp 5
State
0 1 2 3 4 5
Inputs a 1 1 3 1 5 1
b 0 2 0 4 0 4

Chapter 32: Slide – 12

 FSA for Matching abbaa
string matching
rabin karp a a
rabin karp 2
rabin karp 3 a b
rabin karp 4
fsa 1
b 0 a 1 b 2 b 3 a 4 a 5
fsa 2
fsa 3
fsa 4
b b
fsa 5
fsa 6
⊲ fsa 7
Transition Function
kmp 1
kmp 2
kmp 3
kmp 4 State
kmp 5
0 1 2 3 4 5
Inputs a 1 1 1 4 5 1
b 0 2 3 0 2 2

Chapter 32: Slide – 13

 Knuth-Morris-Pratt Algorithm
string matching The Knuth-Morris-Pratt algorithm efficiently
rabin karp
rabin karp 2 implements finite state automatons. It is based on
rabin karp 3
rabin karp 4 computing a prefix function:
fsa 1
fsa 2
fsa 3 π[q] = max{k : k < q and Pk is a suffix of Pq }
fsa 4
fsa 5
fsa 6
fsa 7
where 0 ≤ k < q ≤ m and Pk = P [1 . . . k] and
⊲ kmp 1
Pq = P [1 . . . q]
kmp 2
kmp 3
kmp 4
kmp 5 Pk is a suffix of Pq if Pk = P [q − k + 1..q]

Chapter 32: Slide – 14

 Computing the Prefix Function
string matching
rabin karp
rabin karp 2
Compute-Prefix-Function(P )
rabin karp 3
rabin karp 4
m ← P.length
fsa 1
fsa 2
π[1] ← 0
fsa 3
fsa 4
k←0
fsa 5
fsa 6
for q ← 2 to m
fsa 7
kmp 1
while k > 0 and P [k + 1] 6= P [q]
⊲ kmp 2 k ← π[k]
kmp 3
kmp 4 if P [k + 1] = P [q]
kmp 5
k ← k+1
π[q] ← k
return π

Chapter 32: Slide – 15

 Prefix Function Analysis
string matching
rabin karp Running time is Θ(m). Count changes to k.
rabin karp 2
rabin karp 3 π[k] < k, so k ← π[k] decreases k.
rabin karp 4
fsa 1 k is incremented m − 1 times and k ≥ 0, so
fsa 2
fsa 3 k can be decreased at most m − 1 times.
fsa 4
fsa 5
fsa 6
fsa 7
If P [q] = P [k + 1], then π[q] = k + 1.
kmp 1
kmp 2
If P [q] 6= P [k + 1], then check π[k] next
⊲ kmp 3 because Pπ[k] is a suffix of both Pk and Pq−1 .
kmp 4
kmp 5

Chapter 32: Slide – 16

 KMP Matcher
string matching
rabin karp
rabin karp 2
KMP-Matcher(T, P )
rabin karp 3
rabin karp 4
n ← T.length, m ← P.length
fsa 1
fsa 2
π ← Compute-Prefix-Function(P )
fsa 3
fsa 4
q ← 0 // q is the state of the FSA.
fsa 5
fsa 6
for i ← 1 to n
fsa 7
kmp 1
while q > 0 and P [q + 1] 6= T [i]
kmp 2
kmp 3
q ← π[q]
⊲ kmp 4 if P [q + 1] = T [i] then q ← q + 1
kmp 5
if q = m
pattern occurs with shift i − m
q ← π[q]

Chapter 32: Slide – 17

 KMP Matcher Analysis
string matching
rabin karp Running time is O(n+m). Count changes to q.
rabin karp 2
rabin karp 3 π[q] < q, so q ← π[q] decreases q.
rabin karp 4
fsa 1 q is incremented O(n) times and q ≥ 0, so
fsa 2
fsa 3 q can be decreased at most O(n) times.
fsa 4
fsa 5
fsa 6
fsa 7
Show correctness of computation.
kmp 1
kmp 2
Loop invariant is Pq = T [i − q . . . i − 1].
kmp 3
kmp 4
This is true before the first iteration.
⊲ kmp 5 In while loop,
If P [q + 1] = T [i], then q is incremented.
If P [q] 6= T [i], then check π[q] next because Pπ[q]
is also a suffix of Ti−1 .

Chapter 32: Slide – 18