1. CPSC 311, Fall 2009: Dynamic Programming 1 CPSC 311 Analysis of Algorithms Dynamic Programming Prof. Jennifer Welch Fall 2009

2. CPSC 311, Fall 2009: Dynamic Programming2 Drawback of Divide & Conquer Sometimes can be inefficient Fibonacci numbers: F 0 = 0 F 1 = 1 F n = F n-1 + F n-2 for n > 1 Sequence is 0, 1, 1, 2, 3, 5, 8, 13, …

3. CPSC 311, Fall 2009: Dynamic Programming3 Computing Fibonacci Numbers Obvious recursive algorithm: Fib(n): if n = 0 or 1 then return n else return (F(n-1) + Fib(n-2))

4. CPSC 311, Fall 2009: Dynamic Programming4 Recursion Tree for Fib(5) Fib(5) Fib(4) Fib(3) Fib(2) Fib(1) Fib(2)Fib(1) Fib(0)Fib(1)Fib(0) Fib(1)Fib(0)

5. CPSC 311, Fall 2009: Dynamic Programming5 How Many Recursive Calls? If all leaves had the same depth, then there would be about 2 n recursive calls. But this is over-counting. However with more careful counting it can be shown that it is  ((1.6) n ) Exponential!

6. CPSC 311, Fall 2009: Dynamic Programming6 Save Work Wasteful approach - repeat work unnecessarily Fib(2) is computed three times Instead, compute Fib(2) once, store result in a table, and access it when needed

7. CPSC 311, Fall 2009: Dynamic Programming7 More Efficient Recursive Alg F[0] := 0; F[1] := 1; F[n] := Fib(n); Fib(n): if n = 0 or 1 then return F[n] if F[n-1] = NIL then F[n-1] := Fib(n-1) if F[n-2] = NIL then F[n-2] := Fib(n-2) return (F[n-1] + F[n-2]) computes each F[i] only once

8. CPSC 311, Fall 2009: Dynamic Programming8 1 0 Example of Memoized Fib NIL F 012345012345 1 2 3 5 Fib(5) Fib(4) Fib(3) Fib(2) returns 0+1 = 1fills in F[2] with 1, returns 1+1 = 2 fills in F[3] with 2, returns 2+1 = 3 fills in F[4] with 3, returns 3+2 = 5

9. CPSC 311, Fall 2009: Dynamic Programming9 Get Rid of the Recursion Recursion adds overhead extra time for function calls extra space to store information on the runtime stack about each currently active function call Avoid the recursion overhead by filling in the table entries bottom up, instead of top down.

10. CPSC 311, Fall 2009: Dynamic Programming10 Subproblem Dependencies Figure out which subproblems rely on which other subproblems Example: F0 F1 F2 F3 … Fn-2 Fn-1 Fn

11. CPSC 311, Fall 2009: Dynamic Programming11 Order for Computing Subproblems Then figure out an order for computing the subproblems that respects the dependencies: when you are solving a subproblem, you have already solved all the subproblems on which it depends Example: Just solve them in the order F 0, F 1, F 2, F 3,…

12. CPSC 311, Fall 2009: Dynamic Programming12 DP Solution for Fibonacci Fib(n): F[0] := 0; F[1] := 1; for i := 2 to n do F[i] := F[i-1] + F[i-2] return F[n] Can perform application-specific optimizations e.g., save space by only keeping last two numbers computed

13. CPSC 311, Fall 2009: Dynamic Programming13 Matrix Chain Order Problem Multiplying non-square matrices: A is n x m, B is m x p AB is n x p whose (i,j) entry is ∑a ik b kj Computing AB takes nmp scalar multiplications and n(m-1)p scalar additions (using basic algorithm). Suppose we have a sequence of matrices to multiply. What is the best order? must be equal

14. CPSC 311, Fall 2009: Dynamic Programming14 Why Order Matters Suppose we have 4 matrices: A, 30 x 1 B, 1 x 40 C, 40 x 10 D, 10 x 25 ((AB)(CD)) : requires 41,200 mults. (A((BC)D)) : requires 1400 mults.

15. CPSC 311, Fall 2009: Dynamic Programming15 Matrix Chain Order Problem Given matrices A 1, A 2, …, A n, where A i is d i-1 x d i : [1] What is minimum number of scalar mults required to compute A 1 · A 2 ·… · A n ? [2] What order of matrix multiplications achieves this minimum? Focus on [1]; see textbook for how to do [2]

16. CPSC 311, Fall 2009: Dynamic Programming16 A Possible Solution Try all possibilities and choose the best one. Drawback is there are too many of them (exponential in the number of matrices to be multiplied) Need to be more clever - try dynamic programming!

17. CPSC 311, Fall 2009: Dynamic Programming17 Step 1: Develop a Recursive Solution Define M(i,j) to be the minimum number of mults. needed to compute A i · A i+1 ·… · A j Goal: Find M(1,n). Basis: M(i,i) = 0. Recursion: How to define M(i,j) recursively?

18. CPSC 311, Fall 2009: Dynamic Programming18 Defining M(i,j) Recursively Consider all possible ways to split A i through A j into two pieces. Compare the costs of all these splits: best case cost for computing the product of the two pieces plus the cost of multiplying the two products Take the best one M(i,j) = min k (M(i,k) + M(k+1,j) + d i-1 d k d j )

19. CPSC 311, Fall 2009: Dynamic Programming19 Defining M(i,j) Recursively (A i ·…· A k )·(A k+1 ·… · A j ) P1P1 P2P2 minimum cost to compute P 1 is M(i,k) minimum cost to compute P 2 is M(k+1,j) cost to compute P 1 · P 2 is d i-1 d k d j

20. CPSC 311, Fall 2009: Dynamic Programming20 Step 2: Find Dependencies Among Subproblems 12345 10 2n/a0 3 0 4 0 5 0 GOAL! M: computing the pink square requires the purple ones: to the left and below.

21. CPSC 311, Fall 2009: Dynamic Programming21 Defining the Dependencies Computing M(i,j) uses everything in same row to the left: M(i,i), M(i,i+1), …, M(i,j-1) and everything in same column below: M(i,j), M(i+1,j),…,M(j,j)

22. CPSC 311, Fall 2009: Dynamic Programming22 Step 3: Identify Order for Solving Subproblems Recall the dependencies between subproblems just found Solve the subproblems (i.e., fill in the table entries) this way: go along the diagonal start just above the main diagonal end in the upper right corner (goal)

23. CPSC 311, Fall 2009: Dynamic Programming23 Order for Solving Subproblems 12345 10 2n/a0 3 0 4 0 5 0 M:

24. CPSC 311, Fall 2009: Dynamic Programming24 Pseudocode for i := 1 to n do M[i,i] := 0 for d := 1 to n-1 do // diagonals for i := 1 to n-d to // rows w/ an entry on d-th diagonal j := i + d // column corresponding to row i on d-th diagonal M[i,j] := infinity for k := 1 to j-1 to M[i,j] := min(M[i,j], M[i,k]+M[k+1,j]+d i-1 d k d j ) endfor running time O(n 3 ) pay attention here to remember actual sequence of mults.

25. CPSC 311, Fall 2009: Dynamic Programming25 Example M: 1234 1012007001400 2n/a0400650 3n/a 010,000 4n/a 0 1: A is 30x1 2: B is 1x40 3: C is 40x10 4: D is 10x25

26. CPSC 311, Fall 2009: Dynamic Programming26 Keeping Track of the Order It's fine to know the cost of the cheapest order, but what is that cheapest order? Keep another array S and update it when computing the minimum cost in the inner loop After M and S have been filled in, then call a recursive algorithm on S to print out the actual order

27. CPSC 311, Fall 2009: Dynamic Programming27 Modified Pseudocode for i := 1 to n do M[i,i] := 0 for d := 1 to n-1 do // diagonals for i := 1 to n-d to // rows w/ an entry on d-th diagonal j := i + d // column corresponding to row i on d-th diagonal M[i,j] := infinity for k := 1 to j-1 to M[i,j] := min(M[i,j], M[i,k]+M[k+1,j]+d i-1 d k d j ) if previous line changed value of M[i,j] then S[i,j] := k endfor keep track of cheapest split point found so far: between A k and A k+1 )

28. CPSC 311, Fall 2009: Dynamic Programming28 Example M: 1234 1012007001400 2n/a0400650 3n/a 010,000 4n/a 0 1: A is 30x1 2: B is 1x40 3: C is 40x10 4: D is 10x25 1 2 3 1 3 1 S:

29. CPSC 311, Fall 2009: Dynamic Programming29 Using S to Print Best Ordering Call Print(S,1,n) to get the entire ordering. Print(S,i,j): if i = j then output "A" + i //+ is string concat else k := S[i,j] output "(" + Print(S,i,k) + Print(S,k+1,j) + ")"

30. CPSC 311, Fall 2009: Dynamic Programming30 Example S: 1234 1n/a111 2 23 3 3 4 <<draw recursion tree on board>>