1. Week 2: Greedy Algorithms
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

2. CS4335 Design and Analysis of Algorithms/WANG Lusheng
Greedy Algorithm
A technique to solve problems:
always makes the best choice at the moment (local optimal).
Hopefully, a series of local optimal choices will lead to a globally optimal solution.
Greedy algorithms yield optimal solutions for many (but not all) problems.
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

3. CS4335 Design and Analysis of Algorithms/WANG Lusheng
Knapsack problem:
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

4. CS4335 Design and Analysis of Algorithms/WANG Lusheng
Knapsack problem:
Input: n items: (w1, v1), …, (wn, vn) and a number W
the i-th item’s is worth vi dollars with weight wi.
at most W pounds to be carried.
Output: Some items which are most valuable and with total weight at most W.
Two versions:
0-1 Knapsack: Items cannot be divided into pieces
fractional knapsack: Items can be divided into pieces
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

5. Fractional Knapsack problem
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

6. Fractional Knapsack problem
How many units to take? =7
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

7. CS4335 Design and Analysis of Algorithms/WANG Lusheng
Another Example
a b c d e f 11 kg 3kg 4kg 58kg 8kg 88kg $3 $6 $35 $8 $28 $66
W=20kg
First, compute unit prices
$ $ $8.75 $ $3.50 $0.75
Sort the items according to prices, and then take items until the bucket is full.
(c, 4kg), (e, 8kg), (b, 3kg), (f, 5kg)
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

8. fractional Knapsack Algorithm
Greedy on the unit price: vi/wi.
Each time, take the item with maximum vi/wi.
If exceeds W, take fractions of the item.
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

9. Proof: (correctness, the hard part)
Let X = i1, i2, …ik be an optimal set of items taken.
We just need to prove that our solution is as good as X.
Consider the item with highest unit price, assume it is item j: (vj, wj)
if j is not used in X, get rid of some items with total weight wj and add item j.
Total value is increased. Why?
It implies that item j must be in X.
Repeat the process, X is changed to contain all items selected by greedy WITHOUT decreasing the total value taken.
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

10. 0-1 knapsack problem cannot be solved optimally by greedy
Counter example: (moderate part)
We have three items:
a:($10, 10kg), b:($10, 10kg), c:($12, 11kg)
W=20kg
Greedy approach will take c.
Total value is $12.
Optimal solution will take a and b
Total value is $20.
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

11. Interval Scheduling

12. Interval Scheduling Interval scheduling. a b c d e f g h Time
Job j starts at sj and finishes at fj.
Two jobs compatible if they don't overlap.
Goal: find maximum subset of mutually compatible jobs. a b c
Activity selection = interval scheduling
OPT = B, E, H
Note: smallest job (C) is not in any optimal solution, job that starts first (A) is not in any optimal solution. d e f g h
Time 1 2 3 4 5 6 7 8 9 10 11

13. Ideas for Interval Scheduling
What are possible rules for next Interval?
Choose the interval that starts earliest.
Rationale: start using the resource as soon as possible.
Choose the smallest interval.
Rationale: try to t in lots of small jobs.
Choose the interval that overlaps with the fewest remaining intervals.
Rationale: keep our options open and eliminate as few intervals as possible.
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

14. CS4335 Design and Analysis of Algorithms/WANG Lusheng
Rules That Don’t Work
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

15. CS4335 Design and Analysis of Algorithms/WANG Lusheng
Rules That Don’t Work
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

16. CS4335 Design and Analysis of Algorithms/WANG Lusheng
Rules That Don’t Work
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

17. Optimal Greedy Algorithm
Choose the interval with the earliest finishing time.
Rationale: ensure we have as much of the resource left as possible.
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

18. CS4335 Design and Analysis of Algorithms/WANG Lusheng
Example:
Jobs (s, f): (0, 10), (3, 4), (2, 8), (1, 5), (4, 5), (4, 8), (5, 6) (7,9).
Sorting based on fi:
(3, 4) (1, 5), (4, 5) (5, 6) (4,8) (2,8) (7, 9)(0,10).
Selecting jobs:
(3, 4), remove (1, 5), (2, 8), (0, 10)
(4, 5), remove (4, 8)
(5, 6), remove nothing
(7, 9), remove nothing
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

19. Interval Scheduling: Greedy Algorithm
Greedy algorithm. Consider jobs in increasing order of finish time. Take each job provided it's compatible with the ones already taken.
Implementation. O(n log n).
Remember job j* that was added last to A.
Job j is compatible with A if sj  fj*.
Sort jobs by finish times so that f1  f2  ...  fn.
A  
for j = 1 to n {
if (job j compatible with A)
A  A  {j} }
return A
jobs selected

20. CS4335 Design and Analysis of Algorithms/WANG Lusheng
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

21. How to Prove Optimality
How can we prove the schedule returned is optimal?
Let A be the schedule returned by this algorithm.
Let OPT be some optimal solution.
Might be hard to show that A = OPT, instead we need only to show that |A| = |OPT|.
Note the distinction: instead of proving directly that a choice of intervals A is the same as an optimal choice, we prove that it has the same number of intervals as an optimal. Therefore, it is optimal.
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

22. Interval Scheduling: Analysis
Theorem. Greedy algorithm is optimal.
Pf. (by contradiction)
Assume greedy is not optimal, and let's see what happens.
Let i1, i2, ... ik denote set of jobs selected by greedy.
Let j1, j2, ... jm denote set of jobs in the optimal solution with i1 = j1, i2 = j2, ..., ir = jr for the largest possible value of r.
job ir+1 finishes before jr+1
Greedy: i1 i1 ir
ir+1
OPT: j1 j2 jr
jr+1
. . .
why not replace job jr+1 with job ir+1?

23. Interval Scheduling: Analysis
Theorem. Greedy algorithm is optimal.
Pf. (by contradiction)
Assume greedy is not optimal, and let's see what happens.
Let i1, i2, ... ik denote set of jobs selected by greedy.
Let j1, j2, ... jm denote set of jobs in the optimal solution with i1 = j1, i2 = j2, ..., ir = jr for the largest possible value of r.
job ir+1 finishes before jr+1
Greedy: i1 i1 ir
ir+1
OPT: j1 j2 jr
ir+1
. . .
solution still feasible and optimal, but contradicts maximality of r.

24. Interval Scheduling : ExampleB C A E D F G H
Time 1 2 3 4 5 6 7 8 9 10 11 1 2 3 4 5 6 7 8 9 10 11

25. Interval Scheduling B C A E D F G H Time 1 2 3 4 5 6 7 8 9 10 11 B 1 21 2 3 4 5 6 7 8 9 10 11 B 1 2 3 4 5 6 7 8 9 10 11

26. Interval Scheduling B C A E D F G H Time 1 2 3 4 5 6 7 8 9 10 11 B C 11 2 3 4 5 6 7 8 9 10 11 B C 1 2 3 4 5 6 7 8 9 10 11

27. Interval Scheduling B C A E D F G H Time 1 2 3 4 5 6 7 8 9 10 11 A B 11 2 3 4 5 6 7 8 9 10 11 A B 1 2 3 4 5 6 7 8 9 10 11

28. Interval Scheduling B C A E D F G H Time 1 2 3 4 5 6 7 8 9 10 11 B E 11 2 3 4 5 6 7 8 9 10 11 B E 1 2 3 4 5 6 7 8 9 10 11

29. Interval Scheduling B C A E D F G H Time 1 2 3 4 5 6 7 8 9 10 11 B D E1 2 3 4 5 6 7 8 9 10 11 B D E 1 2 3 4 5 6 7 8 9 10 11

30. Interval Scheduling B C A E D F G H Time 1 2 3 4 5 6 7 8 9 10 11 B E F1 2 3 4 5 6 7 8 9 10 11 B E F 1 2 3 4 5 6 7 8 9 10 11

31. Interval Scheduling B C A E D F G H Time 1 2 3 4 5 6 7 8 9 10 11 B E G1 2 3 4 5 6 7 8 9 10 11 B E G 1 2 3 4 5 6 7 8 9 10 11

32. Interval Scheduling B C A E D F G H Time 1 2 3 4 5 6 7 8 9 10 11 B E H1 2 3 4 5 6 7 8 9 10 11 B E H 1 2 3 4 5 6 7 8 9 10 11

33. Interval Partitioning

34. Interval Partitioning
Lecture j starts at sj and finishes at fj.
Goal: find minimum number of classrooms to schedule all lectures so that no two occur at the same time in the same room.
Ex: This schedule uses 4 classrooms to schedule 10 lectures. e j c d g b h a f i 9
9:30 10
10:30 11
11:30 12
12:30 1
1:30 2
2:30 3
3:30 4
4:30
Time

35. Interval Partitioning
Lecture j starts at sj and finishes at fj.
Goal: find minimum number of classrooms to schedule all lectures so that no two occur at the same time in the same room.
Ex: This schedule uses only 3. c d f j b g i a e h 9
9:30 10
10:30 11
11:30 12
12:30 1
1:30 2
2:30 3
3:30 4
4:30
Time

36. Interval Partitioning: Lower Bound on Optimal Solution
Def. The depth of a set of open intervals is the maximum number that contain any given time.
Key observation. Number of classrooms needed  depth.
Ex: Depth of schedule below = 3  schedule below is optimal.
Q. Does there always exist a schedule equal to depth of intervals?
a, b, c all contain 9:30 c d f j b g i a e h 9
9:30 10
10:30 11
11:30 12
12:30 1
1:30 2
2:30 3
3:30 4
4:30
Time

37. CS4335 Design and Analysis of Algorithms/WANG Lusheng
Greedy Algorithm: c d g j b f i
Depth: The maximum No. of jobs required at any time. a e h
Depth:3
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

38. Interval Partitioning: Greedy Algorithm
Greedy algorithm. Consider lectures in increasing order of start time: assign lecture to any compatible classroom.
Implementation. O(n log n).
For each classroom k, maintain the finish time of the last job added.
Keep the classrooms in a priority queue.
Sort intervals by starting time so that s1  s2  ...  sn.
d  0
for j = 1 to n {
if (lecture j is compatible with some classroom k)
schedule lecture j in classroom k
else
allocate a new classroom d + 1
schedule lecture j in classroom d + 1
d  d + 1 }
number of allocated classrooms

39. Interval Partitioning: Greedy Analysis
Observation. Greedy algorithm never schedules two incompatible lectures in the same classroom.
Theorem. Greedy algorithm is optimal.
Pf.
Let d = number of classrooms that the greedy algorithm allocates.
Classroom d is opened because we needed to schedule a job, say j, that is incompatible with all d-1 other classrooms.
Since we sorted by start time, all these incompatibilities are caused by lectures that start no later than sj.
Thus, we have d lectures overlapping at time sj + .
Key observation  all schedules use  d classrooms.
ddepth

40. Scheduling to Minimize Lateness

41. Solving end of Semester blues
Term paper
Exam study
Party!
CS4335 HW
Max lateness = 2
Project 2
Party!
Exam study
CS4335 HW
Term paper
Project
Monday
Tuesday
Wednesday
Thursday
Friday

42. Scheduling to Minimizing Lateness
Minimizing lateness problem.
Single resource processes one job at a time.
Job j requires tj units of processing time and is due at time dj.
If j starts at time sj, it finishes at time fj = sj + tj.
Lateness: j = max { 0, fj - dj }.
Goal: schedule all jobs to minimize maximum lateness L = max j.
Ex: dj 6 tj 3 1 8 2 9 4 14 5 15
Note: finding a maximal size subset of jobs that meet deadline is NP-hard.
lateness = 2
lateness = 0
max lateness = 6
d3 = 9
d2 = 8
d6 = 15
d1 = 6
d5 = 14
d4 = 9 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

43. Minimizing Lateness: Greedy Algorithms
Greedy template. Consider jobs in some order.
[Shortest processing time first] Consider jobs in ascending order of processing time tj.
[Earliest deadline first] Consider jobs in ascending order of deadline dj.
[Smallest slack] Consider jobs in ascending order of slack dj - tj.

44. Minimizing Lateness: Greedy Algorithms
Greedy template. Consider jobs in some order.
[Shortest processing time first] Consider jobs in ascending order of processing time tj.
[Smallest slack] Consider jobs in ascending order of slack dj - tj.
100 1 10 2 tj dj
counterexample
l2=0, l1=0
l1=0, l2=1 10 11
l1=0, l2=1 2 1 10
l1=9, l2=0 tj 1 11 dj
counterexample

45. CS4335 Design and Analysis of Algorithms/WANG Lusheng
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

46. Minimizing Lateness: Greedy Algorithm
Greedy algorithm. Earliest deadline first.
Sort n jobs by deadline so that d1  d2  …  dn
t  0
for j = 1 to n
Assign job j to interval [t, t + tj]
sj  t, fj  t + tj
t  t + tj
output intervals [sj, fj] dj 6 tj 3 1 8 2 9 4 14 5 15
max lateness = 1
d5 = 14
d2 = 8
d6 = 15
d1 = 6
d4 = 9
d3 = 9 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

47. Minimizing Lateness: No Idle Time
Observation. There exists an optimal schedule with no idle time.
Observation. The greedy schedule has no idle time.
d = 4
d = 6
d = 12 1 2 3 4 5 6 7 8 9 10 11
d = 4
d = 6
d = 12 1 2 3 4 5 6 7 8 9 10 11
"machine not working for some reason, yet work still to be done"

48. Minimizing Lateness: Inversions
Def. An inversion in schedule S is a pair of jobs i and j such that: di < dj but j scheduled before i.
Observation. Greedy schedule has no inversions.
Observation. If a schedule (with no idle time) has an inversion, it has one with a pair of inverted jobs scheduled consecutively.
inversion
before swap j i
Recall: jobs sorted in ascending order of due dates

49. Minimizing Lateness: Inversions
Def. An inversion in schedule S is a pair of jobs i and j such that: di < dj but j scheduled before i.
Claim. Swapping two adjacent, inverted jobs reduces the number of inversions by one and does not increase the max lateness.
Pf. Let  be the lateness before the swap, and let  ' be it afterwards.
'k = k for all k  i, j
'i  i
If job j is late:
inversion fi
before swap j i
after swap i j
f'j

51. CS4335 Design and Analysis of Algorithms/WANG Lusheng
di<dj
We check every pair of consecutive numbers, if there is not inversion, then the whole sequenc has no inversion.
n1n2 … nk
Example: 1, 2, 3, 4, 5, 6, 7, 8, 9
1, 2, 6, 7, 3, 4, 5, 8, 9
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

52. CS4335 Design and Analysis of Algorithms/WANG Lusheng
2019/5/14
CS4335 Design and Analysis of Algorithms/WANG Lusheng

53. Greedy Analysis Strategies
Greedy algorithm stays ahead. Show that after each step of the greedy algorithm, its solution is at least as good as any other algorithm's.
Exchange argument. Gradually transform any solution to the one found by the greedy algorithm without hurting its quality.
Structural. Discover a simple "structural" bound asserting that every possible solution must have a certain value. Then show that your algorithm always achieves this bound.
myopic
greed: "at every iteration, the algorithm chooses the best morsel it can swallow, without worrying about the future"
Objective function. Does not explicitly appear in greedy algorithm!
Hard, if not impossible, to precisely define "greedy algorithm."