Algorithms and Theory of Computation. Lecture 9: Dynamic Programming

Transcription

1 Algorithms and Theory of Computation Lecture 9: Dynamic Programming Xiaohui Bei MAS 714 September 10, 2018 Nanyang Technological University MAS 714 September 10, / 21

2 Recursion in Algorithm Design Divide and Conquer: problem reduced to multiple independent sub-problems that are solved separately. Conquer step puts together solution for bigger problem. Examples: merge sort, quick sort, closest pairs, etc. Dynamic Programming: problem reduced to multiple (typically) dependent or overlapping sub-problems. Use memoization to avoid recomputation of common solutions leading to iterative bottom-up algorithm. Nanyang Technological University MAS 714 September 10, / 21

3 Fibonacci Numbers Fibonacci Numbers are defined by recurrence: F(n) = F(n 1) + F(n 2) and F(0) = 0, F(1) = 1. These numbers have many interesting and amazing properties. A journal The Fibonacci Quarterly F(n) = φn (1 φ) n 5, where φ is the golden ratio (1 + 5)/ lim n F(n + 1)/F(n) = φ Nanyang Technological University MAS 714 September 10, / 21

4 Fibonacci Numbers Fibonacci Numbers Given n, compute F(n). First try: Algorithm: Fib(n): if n = 0 then return 0 else if n = 1 then return 1 else return Fib(n 1) + Fib(n 2) Nanyang Technological University MAS 714 September 10, / 21

5 Fibonacci Numbers Running time analysis: and T(0) = T(1) = 1 T(n) = T(n 1) + T(n 2) + 1 T(n) = Θ(φ n ), roughly same as F(n) The number of additions is exponential in n. Can we do better? Nanyang Technological University MAS 714 September 10, / 21

6 Recursion with memoization An iterative algorithm Algorithm: FibIter(n): if n = 0 then return 0 else if n = 1 then return 1 F[0] = 0, F[1] = 1; for i = 2 to n do F[i] = F[i 1] + F[i 2]; return F[n] Running time: O(n) additions. Nanyang Technological University MAS 714 September 10, / 21

7 Intuition Recursive algorithm is doing the same computation again and again. Iterative algorithm is storing computed values and building bottom up the final value. memoization. Dynamic Programming Finding a recursion that can be effectively/efficiently memorized. Nanyang Technological University MAS 714 September 10, / 21

8 Automatic memoization Can we convert recursive algorithm into an efficient algorithm without explicitly doing an iterative algorithm? Algorithm: Fib(n): if n = 0 then return 0 else if n = 1 then return 1 if Fib(n) was previously computed then return stored value of Fib(n) else return Fib(n 1) + Fib(n 2) How do we keep track of previously computed values? explicitly implicitly (via data structure) Nanyang Technological University MAS 714 September 10, / 21

9 Polynomial Time Issues Is the iterative algorithm a polynomial time algorithm? Input is n and hence input size is Θ(log n). So no. Output is F(n) and output size is Θ(n). Output size is exponential in input size = no polynomial time algorithm possible! Nanyang Technological University MAS 714 September 10, / 21

10 Longest Common Subsequence (LCS) A sequence is an ordered list a 1, a 2,..., a n. The length of a sequence is the number of elements in the list. a i1,..., a ik is a subsequence of a 1,..., a n if 1 i 1 < i 2 < < i k n. Longest Common Subsequence (LCS) Given two sequence X = x 1, x 2,..., x m and Y = y 1, y 2,..., y n, find the length of a longest sequence Z that is a subsequence of both X and Y. Example Sequence: X = 6, 3, 5, 2, 7, 8, 1, 9 Subsequences: 5, 2, 1 and 3, 7, 1, 9, etc Applications: Unix diff, speech recognition, computational biology,... Nanyang Technological University MAS 714 September 10, / 21

11 Naive Enumeration Assume X and Y are contained in two arrays X[1..m], Y[1..n] Algorithm: LCSNaive(X[1..m], Y[1..n]): ans= 0; foreach subsequence Z of X do if Z is also a subsequence of Y and Z > ans then ans = Z ; return ans Running time: (n + m)2 m 2 m subsequences of X and O(m + n) time to check if Z is a subsequence of Y. Nanyang Technological University MAS 714 September 10, / 21

12 Problem Structure Let Z[1..k] be the LCS of sequences X[1..m] and Y[1..n]. 1 If x m = y n, then z k = x m = y n, and Z[1..k 1] is LCS of X[1..m 1] and Y[1..n 1] why? proof by greedy 2 If x m y n and z k x m, then Z is LCS of X[1..m 1] and Y[1..n] 3 If x m y n and z k y n, then Z is LCS of X[1..m] and Y[1..n 1] Let C[i, j] denote the length of an LCS of sequences X[1..i] and Y[1..j]. 0 if i = 0 or j = 0 C[i, j] = C[i 1, j 1] + 1 if x i = y j max{c[i, j 1], C[i 1, j]} otherwise Nanyang Technological University MAS 714 September 10, / 21

13 Dynamic Programming for LCS Algorithm: LCS(X[1..m], Y[1..n]): for i = 0 to m do C[i, 0] = 0; for j = 0 to n do C[0, j] = 0; for i = 1 to m do for j = 1 to n do if X[i] = Y[j] then C[i, j] = C[i 1, j 1] + 1; else C[i, j] = max{c[i, j 1], C[i 1, j]}; return C[m, n] Running time: O(mn) Space: O(mn) Nanyang Technological University MAS 714 September 10, / 21

14 LCS: the Sequence How to compute the LCS? Create another array A of arrows during the computation. x i = y j : arrow goes left and up x i y j : C[i, j] = C[i, j 1]: arrow goes left C[i, j] = C[i 1, j]: arrow goes up Construct LCS: follow the arrow starting at C[m, n]. arrows are at elements of the LCS Nanyang Technological University MAS 714 September 10, / 21

15 LCS Example B A B A C C A B C LCS(BABAC, CABC) = ABC Nanyang Technological University MAS 714 September 10, / 21

16 Dynamic Programming: Summary 1 Find a smart recursion for the problem in which the number of distinct subproblems is small. polynomial in the original problem size 2 Natural ordering of subproblems from smallest to largest, compute the problems bottom up by storing the intermediate values in an appropriate data structure. 3 The total running time is upper bounded by the number of subproblems, the time to evaluate each subproblem and the space needed to store the value. Nanyang Technological University MAS 714 September 10, / 21

17 Shortest Paths Shortest Paths Problem Given a directed weighted graph G = (V, E) with arbitrary edge lengths. For edge e = (u, v), l(e) = l(u, v) is its length. Given vertices s, t, find a shortest path from s to t. b 10 f 6 a e c d g h Nanyang Technological University MAS 714 September 10, / 21

18 Negative Lengths Dijkstra s algorithm can fail if there are negative edges lengths. 0s 1 y1 5 x z w 1 3 False Assumption: Dijkstra s algorithm is based on the assumption that if s = v 0 v 1 v k = v is a shortest path from s to v, then dist(s, v i ) dist(s, v) for all 1 i < k. Holds true only for non-negative edge lengths. Nanyang Technological University MAS 714 September 10, / 21

19 Negative Cycles A negative cycle is a directed cycle such that the sum of its edge lengths is negative. a e b c d 25 f g 6 h Nanyang Technological University MAS 714 September 10, / 21

20 Shortest Paths and Negative Cycles Lemma 1 If some path from s to t contains a negative cycle, then there does not exist a shortest path from s to t. Proof. If there exists such a cycle, can build a s t path of arbitrarily negative weight by detouring around cycle as many times as desired. Lemma 2 If G has no negative cycles, then there exists a shortest path from s to t that is simple. Proof. Consider a shortest s t path P that uses the fewest number of edges. If P contains a cycle C, can remove portion of P corresponding to C without increasing the total length. Nanyang Technological University MAS 714 September 10, / 21

21 Shortest Path and Negative Cycle Problems Negative Cycle Problem Given a directed weighted graph G = (V, E) with arbitrary edge lengths, find a negative cycle (if one exists). Shortest Path Problem Given a directed weighted graph G = (V, E) with arbitrary edge lengths and no negative cycles, find a shortest path from s to t. Nanyang Technological University MAS 714 September 10, / 21