Presentation is loading. Please wait.

Presentation is loading. Please wait.

Algorithm Design Methods (II) Fall 2003 CSE, POSTECH.

Similar presentations


Presentation on theme: "Algorithm Design Methods (II) Fall 2003 CSE, POSTECH."— Presentation transcript:

1 Algorithm Design Methods (II) Fall 2003 CSE, POSTECH

2 Quick Sort Quicksort can be seen as a variation of mergesort in which front and back are defined in a different way.

3 Quicksort Algorithm Partition anArray into two non-empty parts. – Pick any value in the array, pivot. – small = the elements in anArray < pivot – large = the elements in anArray > pivot – Place pivot in either part, so as to make sure neither part is empty. Sort small and large by recursively calling QuickSort. You could use merge to combine them, but because the elements in small are smaller than elements in large, simply concatenate small and large, and put the result into anArray.

4 Quicksort: Complexity Analysis Like mergesort, a single invocation of quicksort on an array of size p has complexity O(p): – p comparisons = 2*p accesses – 2*p moves (copying) = 4*p accesses Best case: every pivot chosen by quicksort partitions the array into equal-sized parts. In this case quicksort is the same big-O complexity as mergesort – O(n log n)

5 Quicksort: Complexity Analysis Worst case: the pivot chosen is the largest or smallest value in the array. Partition creates one part of size 1 (containing only the pivot), the other of size p-1. n 1n-1 n-21 1 1

6 Quicksort: Complexity Analysis Worst case: There are n-1 invocations of quicksort (not counting base cases) with arrays of size: p = n, n-1, n-2, …, 2 Since each of these does O(p), the total number of accesses is O(n) + O(n-1) + … + O(1) = O(n 2 ) Ironically the worst case occurs when the list is sorted (or near sorted)!

7 Quicksort: Complexity Analysis The average case must be between the best case O(n log n) and the worst case is O(n 2 ). Analysis yields a complex recurrence relation. The average case number of comparisons turns out to be approximately 1.386*n*log n – 2.846*n. Therefore the average case time complexity is O(n log n).

8 Quicksort: Complexity Analysis Best case O(n log n) Worst case O(n2) Average case O(n log n) Note that the quick sort is inferior to insertion sort and merge sort if the list is sorted, nearly sorted, or reverse sorted.

9 Dynamic Programming Sequence of decisions Problem state Principle of optimality

10 Sequence of Decisions As in the greedy method, the solution to a problem is viewed as the result of a sequence of decisions. Unlike the greedy method, decisions are not made in a greedy manner. Examine the decision sequence to see whether an optimal decision sequence contains optimal decision subsequences.

11 Example: Matrix Chain Product M1 X M2 where M1 is n x m and M2 is m x q Then the total number of multiplications is n x m x q When there is a matrix chain product such as MCP(n) = M1 X M2 X M3 X ….. X Mn, Find a sequence of matrix products that results in the least number of matrix multiplications. Example : M1 X M2 X M3 X M4 – Possible sequence : (M1 X (M2 X (M3 X M4))), ((M1 X M2) X (M3 X M4)), (((M1 X M2) X M3) X M4), ((M1 X (M2 X M3)) X M4), (M1 X ((M2 X M3) X M4),

12 Matrix Chain Product Decision based space decomposition – Compute M1 X M2 first then compute with others or – Compute M2…Mn first then compute with M1 Subspace for M1 X M2 – Compute M3… Mn Subspace for M2..Mn – Compute M3.. Mn first then compute with M2 or – Compute M2 X M3 first then compute with others Recursively break down the space by having a decision Final decision is based on the number of computations required (in a recursive implementation). But, subspaces are repeating.  Use bottom-up approach to avoid the repetitive computation

13 Matrix Chain Product Non-recursive computation of MCP – Compute (M1: M2), (M2: M3), ….., (Mn-1: Mn) – Compute (M1:M3), (M2:M4),...., (Mn-2:Mn) using lower computation results Ex: (M1:M3) =min( (M1 X(M2:M3)), ((M1:M2) XM3) ) – Compute (M1:M4), (M2:M5), ….., (Mn-3:Mn) – ….. – Finally compute (M1:M5) = min( (M1X(M2:Mn)), ((M1:M2)X(M3:Mn), …., ((M1:Mn-1)XMn) ) Time Complexity – 1 st : O(n), 2 nd : O((n-1)X2), 3 rd : O((n-3)X3) ….. N-th: O(1X(n-1)) – Total time complexity O(n**2) Space complexity : O(n**2)

14 Back Tracking Systematic way to search for the solution to a problem Begin by defining a solution space for the problem. – Example: Rat in a maze => solution space: all possible paths to the destination 0/1 Knapsack => solution space: all possible 0/1 combinations Organize the solution space so that it can be searched easily. – Graph or tree Search the space in depth-first manner beginning at the start node. There is no more space to search in dfs, backtrack to the previous live node and expand over there. The search terminates when the destination reached or there is no more live node.

15 BackTracking: Example 0/1 Knapsack – n=3, c = 15, w = [8, 6, 9], p = [5, 4, 6] Search space – (0,-,-),..,(0,0,0), (0,0,1), (0,1,0), …., (1,1,1) Search start from (1,-,-) to (0,0,0) (1,1,1) gives a violation of capacity, so backtrack to (1,1,-) and choose (1,1,0) that satisfies the constraints  feasible solution with profit 9 Continue search using other live nodes (0,1,1) is another feasible solution with more profit.  optimal solution with profit 10

16 Time Complexity Exhaustive search – O(2**n) Can speed up the search for an optimal solution by introducing bounding function: “Whether a newly reached node can lead to a solution better than the best found so far.” The solution space needed for the search is the path information from the start node to the current expansion node. O(longest path)

17 Branch and Bound Another systematic way to search for the solution to a problem Each live node becomes E (expansion)-node exactly once. When a node becomes an E-node, all new nodes that can be reached using a single move are generated. Generated nodes that cannot possibly lead to a (optimal) feasible solution are discarded. The remaining nodes are added to the list of live nodes and then one node from the list is selected to become the next E-node. The expansion process is continued until either the answer is found or the list of live nodes become empty. Selection method from the list of live nodes – First-in First-out (BFS) – Least Cost or Max Profit

18 Branch and Bound: Example 0/1 Knapsack – n=3, c = 15, w = [8, 9, 6], p = [5, 6, 4] Search space – (1,-,-),…(0,0,0), (0,0,1), (0,1,0), …., (1,1,1) Search start from (-,-,-) and expand – (1,-,-), (0,-,-): both live nodes. Select (1,-,-) by FIFO selection and expand – (1,1,-), (1,0,-): (1,1,-) is infeasible (bound) but (1,0,-) is feasible Select (0,-,-) by FIFO selection and expand – (0,1,-), (0,0,-): both feasible Select (1,0,-) by FIFO selection and expand – (1,0,1), (1,0,0): (1,0,0) is feasible but less profit – (1,0,1) is feasible solution with profit 9

19 Branch and Bound: Example 0/1 Knapsack – n=3, c = 15, w = [8, 9, 6], p = [5, 4, 6] Select (0,1,-) by FIFO selection and expand – (0,1,1), (0,1,0): (0,1,1) feasible with more profit (10) Select (0,0,-) by FIFO selection and expand – (0,0,1) but less profit.  optimal solution (0,1,1) with profit 10

20 Time Complexity Exhaustive search – O(2**n) : depends on bounding function

21 Supplement slides for the dynamic programming

22 Principle of Optimality An optimal solution satisfies the following property: No matter what the first decision is, the remaining decisions are optimal with respect to the state that results from this decision. Dynamic programming may be used only when the principle of optimality holds.

23 0/1 Knapsack Problem Suppose that decisions are made in the order x 1, x 2, x 3, …, x n. Let x 1 =a 1, x 2 =a 2, …, x n =a n be an optimal solution. If a 1 = 0, then following the first decision the state is (2,c). a 2, a 3 …, a n must be an optimal solution to the knapsack instance given by the state (2,c).

24 x 1 = a 1 = 0 maximize Sigma(i=2…n) p i x i. subject to Sigma(i=2…n) w i x i <= c and x i = 0 or 1 for all i. If not, this instance has a better solution b 2, b 3, …, b n. Sigma(i=2,…n) p i b i > Sigma(i=2…n) p i a i.

25 x 1 = a 1 = 0 x 1 =a 1, x 2 =b 2, x 3 =b 3 …, x n =b n is a better solution to the original instance than is x 1 =a 1, x 2 =a 2, x 3 =a 3 …, x n =a n. So x 1 =a 1, x 2 =a 2, x 3 =a 3 …, x n =a n cannot be an optimal solution … a contradiction with the assumption that it is optimal.

26 x 1 = a 1 = 1 Next, consider the case a 1 = 1. Following the first decision the state is (2,c-w 1 ). a 2, a 3 …, a n must be an optimal solution to the knapsack instance given by the state (2,c- w 1 ).

27 x 1 = a 1 = 1 maximize Sigma(i=2…n) p i x i. subject to Sigma(i=2…n) w i x i <= c-w 1 and x i = 0 or 1 for all i. If not, this instance has a better solution b 2, b 3, …, b n. Sigma(i=2,…n) p i b i > Sigma(i=2…n) p i a i.

28 x 1 = a 1 = 1 x 1 =a 1, x 2 =b 2, x 3 =b 3 …, x n =b n is a better solution to the original instance than is x 1 =a 1, x 2 =a 2, x 3 =a 3 …, x n =a n. So x 1 =a 1, x 2 =a 2, x 3 =a 3 …, x n =a n cannot be an optimal solution … a contradiction with the assumption that it is optimal.

29 0/1 Knapsack Problem Therefore, no matter what the first decision is, the remaining decisions are optimal with respect to the state that results from this decision. The principle of optimality holds and dynamic programming may be applied.

30 Dynamic Programming Recurrence f(n,y) is the value of the optimal solution to the knapsack instance defined by the state (n,y). – Only item n is available. – Available capacity is y. If w n <= y, f(n,y) = p n. If w n > y, f(n,y) = 0.

31 Dynamic Programming Recurrence Suppose that i < n. f(i,y) is the value of the optimal solution to the knapsack instance defined by the state (i,y). – Items i through n are available. – Available capacity is y. Suppose that in the optimal solution for the state (i,y), the first decision is to set x i = 0. From the principle of optimality, it follows that f(i,y) = f(i+1,y).

32 Dynamic Programming Recurrence The only other possibility for the first decision is x i = 1. The case x i = 1 can arise only when y >= w i. From the principle of optimality, it follows that f(i,y) = f(i+1,y-w i ) + p i. Combining the two cases, we get – f(i,y) = f(i+1,y) if y < w i. – f(i,y) = max{f(i+1,y), f(i+1,y-w i )+p[i]} if y >= w i.

33 Recursive Code /** @return f(i,y) **/ private static int f (int i, int y) { if (i == n) return (y < w[n])? 0:p[n]; if (y < w[i]) return f(i+1,y); return max (f(i+1,y), f(i+1, y-w[i])+p[i]); }

34 Recursion Tree

35 Time Complexity Let t(n) be the time required when n items are available. t(0) = t(1) = a, where a is a constant. When n > 1, t(n) <= 2t(n-1) + b, where b is a constant. t(n) = O(2 n ). Solving dynamic programming recurrences recursively can be hazardous to run time.

36 Reducing Run Time

37 Time Complexity Level i of the recurrence tree has up to 2 i-1 nodes. At each such node an f(i,y) is computed. Several nodes may compute the same f(i,y). We can save time by not recomputing already computed f(i,y)s. Save computed f(i,y)s in a dictionary. – Key is (i,y) value. – f(i,y) is computed recursively only when (i,y) is not in the dictionary. – Otherwise, the dictionary value is used.


Download ppt "Algorithm Design Methods (II) Fall 2003 CSE, POSTECH."

Similar presentations


Ads by Google