CS345-Algorithms-II-Lecture-1-CS345-2016.pdf
- 1. Design and Analysis of Algorithms Lecture 1 • Overview of the course • Closest Pair problem 1 https://moodle.cse.iitk.ac.in CS345A Algorithms-II
- 2. Aim of the course To empower each student with the skills to design algorithms • With provable guarantee on correctness. • With provable guarantee on their efficiency. 2
- 3. Algorithm Paradigm Motivation: • Many problems whose algorithms are based on a common approach. ➔ A need of a systematic study of the characteristics of such approaches. Algorithm Paradigms: • Divide and Conquer • Greedy Strategy • Dynamic Programming 3 (advanced) (advanced)
- 4. Maximum Flow Given a network for transporting certain commodity (water/bits) from a designated source vertex 𝒔 and sink vertex 𝒕. Each edge has a certain capacity (max rate per unit time at which commodity can be pumped along that edge), Compute the maximum rate at which we can pump flow from 𝒔 to 𝒕. Constraints: capacity constraint and conservation constraint. 4 𝒔 𝒗 𝒖 𝒙 𝒚 𝒕 2 17 5 6 8 17 4 15 16 7 14
- 5. Miscellaneous • Matching in Graphs Maximum matching, Stable matching • Amortized Analysis A powerful technique to analyse time complexity of algorithms • String Matching • Linear Programming 5
- 6. Last topic on Algorithms • NP Complete problems • Approximation/randomized Algorithms 6
- 8. Data structures • Augmented Binary Search Trees • Range Minima Data structure (optimal size) • Fibonacci Heap 8 : Additional information
- 9. Orthogonal Range searching Problem: Preprocess a set of 𝒏 points so that given any query rectangle, the number of points lying inside it can be reported efficiently. Data structure: size = O(𝒏 log 𝒏), Query = O( log2 𝒏), size = O(𝒏), Query = O( 𝑛), 9 Rectangle A novel application of augmented BST Try to solve it… You can surely do it…☺ Rectangle
- 10. Divide and Conquer 10 A paradigm for Algorithm Design
- 11. An Overview A problem in this paradigm is solved in the following way. 1. Divide the problem instance into two or more instances of the same problem. 2. Solve each smaller instance recursively (base case suitably defined). 3. Combine the solutions of the smaller instances to get the solution of the original instance. 11 This is usually the main nontrivial step in the design of an algorithm using divide and conquer strategy
- 12. Example Problems 1. Merge Sort 2. Multiplication of two 𝒏-bit integers. 3. Counting the number of inversions in an array. 4. Median finding in linear time. 12
- 13. PROBLEM 1 Closest Pair of Points 13
- 14. The Closest Pair Problem 14 𝜹
- 15. Closest Pair of Points Problem Definition: Given a set 𝑷 of 𝒏 > 𝟏 points in plane, compute the pair of points with minimum Euclidean distance. Deterministic algorithms: • O(𝒏𝟐) : Trivial algorithm • O(𝒏 𝐥𝐨𝐠 𝒏) : Divide and Conquer based algorithm 15
- 16. Hint/Tool No. 1 Exercise: What is the maximum number of points that can be placed in a unit square such that the minimum distance is at least 1 ? Answer: 4. 16 1 1 2 A discrete math exercise If there are more than 4 points, at least one of the four small squares will have more than 1 points.
- 17. Hint/Tool No. 2 Question: For which algorithmic problems do we need a suitable data structure ? Answer: If the problem involves “many” operations of same type on a given data. For example, it is worth sorting an array only if there are going to be many search queries on it. Let us see if you can use this principle in today’s class itself ☺ 17 When do we use a data structure ?
- 19. The conquer step 19 𝜹𝑳 𝜹𝑹 Compute closest pair of the left half set Compute closest pair of the right half set Notice 𝜹𝑳 < 𝜹𝑹 for this given instance
- 20. The combine step 20 𝜹𝑳 𝜹𝑹 𝜹𝑳 𝜹𝑳 Which points do we need to focus on for the closest pairs ?
- 21. The combine step 21 𝜹𝑳 𝜹𝑹 𝜹𝑳 𝜹𝑳 But there may still be Θ(𝑛2 ) pairs of points here So what to do ?
- 22. The combine step 22 𝜹𝑳 𝜹𝑹 𝜹𝑳 𝜹𝑳 Focus on a point 𝒑 in left strip. Where do we have to search for the points in the right strip that can form a pair with 𝒑 at distance < 𝜹𝑳 ? 𝒑
- 24. The combine step 24 𝜹𝑳 𝜹𝑹 𝜹𝑳 𝜹𝑳 𝜹𝑳 𝜹𝑳 Only the points lying in these 2 red squares are relevant as far as 𝒑 is concerned. 𝒑 How many points can there be in these 2 red squares each of length𝜹𝑳? Surely not more than 8 (using Hint 1) How to find the points in these red square for point 𝒑 ? It will take O(𝒏) time for a given 𝒑. It is time to use Hint/Tool no. 2. Think for a while before going to the next slide.
- 25. The combine step 25 𝜹𝑳 𝜹𝑹 𝜹𝑳 𝜹𝑳 𝜹𝑳 𝜹𝑳 We need to find points in the 2 red square for every point in the left strip. So build a suitable data structure for points in the right strip so that we can answer such query efficiently for each point in the left strip. What will be the data structure ? An array storing the points of the right strip in increasing order of y-coordinates. 𝜹𝑳 𝜹𝑳 𝜹𝑳 𝜹𝑳
- 26. Divide and Conquer based algorithm CP-Distance(𝑃) { If (| 𝑃 |=1 ) return infinity; { Compute 𝑥-median of 𝑃; (𝑃𝐿, 𝑃𝑅)Split-by-𝑥-median(𝑃); 𝜹𝑳 CP-Distance(𝑃𝐿) ; 𝜹𝑹 CP-Distance(𝑃𝑅) ; 𝜹 min(𝜹𝑳, 𝜹𝑹); 𝑆𝐿 strip of 𝑃𝐿; 𝑆𝑅 strip of 𝑃𝑅; 𝐴 Sorted array of 𝑆𝑅; For each 𝑝 ∈ 𝑆𝐿, 𝒚 y-coordinate of 𝑝; Search 𝐴 for points with y-coordinate within 𝒚 ± 𝜹; Compute distance from 𝑝 to each of these points; Update 𝜹 accordingly; return 𝜹; } 26 Divide step Combine/conquer step 𝑶( 𝐏 log 𝐏) time 𝑶( 𝐏 ) + 2 T(|𝐏|/2) time
- 27. Running time of the algorithm What is the recurrence for running time? T(𝑛) = c 𝑛 log 𝑛 + 2 T(𝑛/2) ➔ T(𝑛) = O( 𝑛 log2𝑛) Theorem: There exists an O( 𝑛 log2𝑛) time algorithm to compute closest pair of 𝑛 points in plane. 27
- 28. Conclusion Homework: 1. Try to improve the running time to O( 𝑛 log 𝑛). Hint: “the code will look similar to that of MergeSort”. 2. Ponder over the data structure for orthogonal range searching. 28
- 29. How does one design an algorithm ? If you wish to find the answer on your own, try to solve the first assignment problem on your own. 29 Without any help from the web Without any help from the your friends
- 30. Assignment 1 Smallest Enclosing circle Problem definition: Given 𝒏 points in a plane, compute the smallest radius circle that encloses all 𝒏 point. 30