June 11, Dynamic Programming( Weighted Interval Scheduling)

Transcription

1 Dynamic Programming( Weighted Interval Scheduling) June 11, 2014

2 Problem Statement: 1 We have a resource and many people request to use the resource for periods of time (an interval of time) 2 Each interval (request) i has a start time s i and finish time f i 3 Each interval (request) i, has a value v i Conditions: the resource can be used by at most one person at a time. we can accept only compatible intervals (requests) (overlap-free). Goal: a set of compatible intervals (requests) with a maximum total value.

3 What if all the values are the same?

4 What if all the values are the same? We start with the one which has the smallest finish time and add it into our solution and remove the ones that have intersection with it and repeat!

5 1 v 1 = v 2 = 4 v 3 = 4 v 4 = 7 v 5 = 2 v 6 = 1

6 1 v 1 = v 2 = 4 v 3 = 4 v 4 = 7 v 5 = 2 v 6 = 1 1 v 1 = v 2 = 4 v 3 = 4 v 4 = 7 v 5 = 5 v 6 = 1

7 We order the intervals (requests) based on their finishing time (non-increasing order of finishing time) : f 1 f 2 f n

8 We order the intervals (requests) based on their finishing time (non-increasing order of finishing time) : f 1 f 2 f n For every interval j let p(j) be the largest index i < j such that intervals i, j do not overlap. p(j) = 0 if there is no interval before j and disjoint from j.

9 We order the intervals (requests) based on their finishing time (non-increasing order of finishing time) : f 1 f 2 f n For every interval j let p(j) be the largest index i < j such that intervals i, j do not overlap. p(j) = 0 if there is no interval before j and disjoint from j. 1 v 1 = 2 p(1) = v 2 = 4 v 3 = 4 v 4 = 7 v 5 = 2 v 6 = 1 p(2) = 0 p(3) = 0 p(4) = 0 p(5) = 3 p(6) = 4

10 Let OPT (j) be the value of the optimal solution considering only intervals from 1 to j (according to their order).

11 Let OPT (j) be the value of the optimal solution considering only intervals from 1 to j (according to their order). We can write a recursive formula to compute OPT (j), Either interval j is in the optimal solution or j is not in the solution.

12 Let OPT (j) be the value of the optimal solution considering only intervals from 1 to j (according to their order). We can write a recursive formula to compute OPT (j), Either interval j is in the optimal solution or j is not in the solution. Therefore : OPT (j) = max{opt (j 1), v j + OPT (p(j))}

13 Let OPT (j) be the value of the optimal solution considering only intervals from 1 to j (according to their order). We can write a recursive formula to compute OPT (j), Either interval j is in the optimal solution or j is not in the solution. Therefore : OPT (j) = max{opt (j 1), v j + OPT (p(j))} Try not to implement using recursive call because the running time would be exponential! Recursive function is easy to implement but time consuming!

14 Dynamic Programming Iterative-Compute-Opt() 1. M[0] := 0; 2. for j = 1 to n 3. M[j] = max{v j + M[p(j)], M[j 1]}

15 Dynamic Programming Iterative-Compute-Opt() 1. M[0] := 0; 2. for j = 1 to n 3. M[j] = max{v j + M[p(j)], M[j 1]} Find-Solution(j) 1. if j = 0 then return 2. else 3. if (v j + M[p(j)] M[j 1]) 4. print j and print 5. Find-Solution(p(j)) 6. else 7. Find-Solution(j-1)

16 Exercise Problem: Given a weighted acyclic digraph D (weight are on the arcs), find a longest weighted path from source node s to sink node t. Source is the only node that has no incoming arc and sink is the only node that has no out-going arc.

17 1 v 1 = 2 p(1) = v 2 = 4 v 3 = 4 v 4 = 7 v 5 = 2 p(2) = 0 p(3) = 0 p(4) = 0 p(5) = 3 6 v 6 = 1 p(6) = 4

18 Subset Sums and Knapsack Problem Subset Sums (Problem Statement): 1 We are given n items {1, 2,..., n} and each item i has weight w i 2 We are also given a bound W Goal: select a subset S of the items so that : 1 Σ i S w i W 2 Σ i S w i is maximized

19 A greedy approach won t work if it is based on picking the biggest value first. Suppose we have a set {W /2 + 1, W /2, W /2} of items. If we choose W /2 + 1 then we can not choose anything else. However the optimal is W /2 + W /2.

20 A greedy approach won t work if it is based on picking the biggest value first. Suppose we have a set {W /2 + 1, W /2, W /2} of items. If we choose W /2 + 1 then we can not choose anything else. However the optimal is W /2 + W /2. A greedy approach won t work if it is based on picking the smallest value first. Suppose we have a set {1, W /2, W /2} of items. If we choose 1 then we can only choose W /2 and nothing more. However the optimal is W /2 + W /2.

21 A greedy approach won t work if it is based on picking the biggest value first. Suppose we have a set {W /2 + 1, W /2, W /2} of items. If we choose W /2 + 1 then we can not choose anything else. However the optimal is W /2 + W /2. A greedy approach won t work if it is based on picking the smallest value first. Suppose we have a set {1, W /2, W /2} of items. If we choose 1 then we can only choose W /2 and nothing more. However the optimal is W /2 + W /2. Exhaustive search! Produce all the subset and check which one satisfies the constraint ( W ) and has maximum size. Running time O(2 n ).

22 Let OPT [i, w] be the optimal maximum value of a set S {1, 2,..., i} where the total value of the items in S is at most w.

23 Let OPT [i, w] be the optimal maximum value of a set S {1, 2,..., i} where the total value of the items in S is at most w. If w i > w then OPT [i, w] = OPT [i 1, w] otherwise

24 Let OPT [i, w] be the optimal maximum value of a set S {1, 2,..., i} where the total value of the items in S is at most w. If w i > w then OPT [i, w] = OPT [i 1, w] otherwise OPT [i, w] = max{opt [i 1, w], w i + OPT [i 1, w w i ]}

25 Subset-Sum(n,W) 1. Array M[0,.., n, 0,.., W ] 2. for ( i = 1 to W ) M[0, i] = 0 3. for ( i = 1 to n ) 4. for ( w = 1 to W ) 5. if (w < w i ) then M[i, w] = M[i 1, w] 6. else 7. if (w i + M[i 1, w w i ] > M[i 1, w]) 8. M[i, w] = w i + M[i 1, w w i ] 9. else M[i, w] = M[i 1, w]

26 Subset-Sum(n,W) 1. Array M[0,.., n, 0,.., W ] 2. for ( i = 1 to W ) M[0, i] = 0 3. for ( i = 1 to n ) 4. for ( w = 1 to W ) 5. if (w < w i ) then M[i, w] = M[i 1, w] 6. else 7. if (w i + M[i 1, w w i ] > M[i 1, w]) 8. M[i, w] = w i + M[i 1, w w i ] 9. else M[i, w] = M[i 1, w] Time complexity O(nW )

27 Exercise Suppose W = 6 and n = 3 and the items of sizes w 1 = w 2 = 2 and w 3 = 3. Fill out matrix M.

28 Problem : We are given n jobs {J 1, J 2,..., J n } where each J i has a processing time p i > 0 ( an integer). We have two identical machines M 1, M 2 and we want to execute all the jobs. Schedule the jobs so that the finish time is minimized.

29 Knapsack Problem Knapsack Problem : 1 We are given n items {1, 2,..., n} and each item i has weight w i 2 Each item has value v i 3 We are also given a bound W Goal: select a subset S of the items so that : 1 Σ i S w i W 2 Σ i S v i is maximized

30 Let OPT [i, w] be the optimal maximum value of a set S {1, 2,..., i} where the total value of the items in S is at most w.

31 Let OPT [i, w] be the optimal maximum value of a set S {1, 2,..., i} where the total value of the items in S is at most w. If w i > w then OPT [i, w] = OPT [i 1, w] otherwise

32 Let OPT [i, w] be the optimal maximum value of a set S {1, 2,..., i} where the total value of the items in S is at most w. If w i > w then OPT [i, w] = OPT [i 1, w] otherwise OPT [i, w] = max{opt [i 1, w], v i + OPT [i 1, w w i ]}