Struktur Data & Algoritme (Data Structures & Algorithms) Analisa Algoritma Denny denny@cs.ui.ac.id Fakultas Ilmu Komputer Universitas Indonesia Semester Genap - 2000/2001 Version 1.0 - Internal Use Only Algoritme Suatu set instruksi yang harus diikuti oleh computer untuk memecahkan suatu masalah. SDA/ALG-AN/DN/V1.0/2 Fundamental Data Types: String, Math, Casting 1
Analisa Algoritme: motivasi perlu diketahui berapa banyak resource (time & space) yang diperlukan oleh sebuah algoritme Menggunakan teknik-teknik untuk mengurangi waktu yang dibutuhkan oleh sebuah algoritme SDA/ALG-AN/DN/V1.0/3 Objectives dapat memperkirakan waktu yang dibutuhkan mempelajari teknik yang secara drastis menurunkan running time mempelajari kerangka matematik yang secara rigorously menggambarkan running time SDA/ALG-AN/DN/V1.0/4 Fundamental Data Types: String, Math, Casting 2
Outline Apa itu analisa algoritme?- what Bagaimana cara untuk analisa/mengukur? - how Notasi Big-Oh Case Study: The Maximum Contiguous Subsequence Sum Problem Algorithm 1: A cubic algorithm Algorithm 2: A quadratic algorithm Algorithm 3: An N log N algorithm Algorithm 4: A linear algorithm SDA/ALG-AN/DN/V1.0/5 Analisa Algoritme: what? Mengukur jumlah sumber daya (time dan space) yang diperlukan oleh sebuah algoritme Waktu yang diperlukan (running time) oleh sebuah algorithm cenderung tergantung pada jumlah input yang diproses. Running time dari sebuah algoritme adalah fungsi dari jumlah inputnya Selalu tidak terikat pada platform, bahasa pemrograman atau bahkan metodologi (mis. Procedural vs OO) SDA/ALG-AN/DN/V1.0/6 Fundamental Data Types: String, Math, Casting 3
Worst, Best and Average Case Worst-case running time is a bound over all inputs of a certain size N. (Guarantee - Pesimistic) Average-case running time is an average over all inputs of a certain size N. (Prediction) Best-case running time is defined by the minimum number of steps taken on any instance of size n. (Optimistic) SDA/ALG-AN/DN/V1.0/7 Worst, Best and Average Case (2) running time worst case average case best case input size n SDA/ALG-AN/DN/V1.0/8 Fundamental Data Types: String, Math, Casting 4
Analisa Algoritme: how? Bagaimana jika kita menggunakan jam? Jumlah waktu yang digunakan bervariasi tergantung pada beberapa faktor lain: kecepatan mesin, sistem operasi, kualitas kompiler, dan bahasa pemrograman. Sehingga kurang memberikan gambaran yang tepat tentang algoritme Notasi O (sering disebut sebagai notasi big-oh ) Digunakan sebagai bahasa untuk membahas efisiensi dari sebuah algoritme: log n, linier, n log n, n 2, n 3,... Contoh: run time algoritme Insertion Sort adalah O(n 2 ) Dari hasil run-time, dapat kita buat grafik dari waktu eksekusi dan jumlah data. SDA/ALG-AN/DN/V1.0/9 Big Oh Sebuah fungsi kubic adalah sebuah fungsi yang suku dominannya (dominant term) adalah sebuah konstan dikalikan dengan n 3. Contoh: 10 n 3 + n 2 + 40 n + 80 n 3 + 1000 n 2 + 40 n + 80 n 3 + 10000 SDA/ALG-AN/DN/V1.0/10 Fundamental Data Types: String, Math, Casting 5
Dominant Term Mengapa hanya suku yang memiliki pangkat tertinggi/dominan saja yang diperhatikan? Contoh: kita estimasi n 3 + 350n 2 + n dengan n 3. Untuk n = 10000 Nilai aktual 1,003,500,010,000 Nilai estimasi 1,000,000,000,000 Kesalahan dalam estimasi 0.35%, dapat diabaikan. Untuk n yang besar, suku dominan lebih mengindikasikan perilaku dari algoritme. Untuk n yang kecil, suku dominan tidak selalu mengindikasikan perilakunya, tetapi program dengan input kecil umumnya berjalan sangat cepat sehingga kita tidak perlu perhatikan. SDA/ALG-AN/DN/V1.0/11 Big Oh: issues Apakah fungsi linier selalu lebih kecil dari fungsi kubic? Untuk n yang kecil, bisa saja fungsi linier > fungsi kubic Tetapi untuk n yang besar, fungsi kubic > fungsi linier Contoh: 10 n 3 + 20 n + 10 1000 n + 100 Mengapa nilai konstan/koefesien pada setiap suku tidak diperhatikan? Nilai konstan/koefesien tidak berarti pada mesin yang berbeda \ Big Oh digunakan untuk merepresentasikan laju pertumbuhan (growth rate) SDA/ALG-AN/DN/V1.0/12 Fundamental Data Types: String, Math, Casting 6
Orde Fungsi Running-Time Function Name c Constant log N Logarithmic log 2 N Log-squared N Linear N log N N log N N 2 Quadratic N 3 Cubic 2 N Exponential N! Factorial SDA/ALG-AN/DN/V1.0/13 Fungsi Running-Time Untuk input yang sedikit, beberapa fungsi lebih cepat dibandingkan dengan yang lain. SDA/ALG-AN/DN/V1.0/14 Fundamental Data Types: String, Math, Casting 7
Fungsi Running-Time (2) Untuk input yang besar, beberapa fungsi runningtime sangat lambat - tidak berguna. SDA/ALG-AN/DN/V1.0/15 Contoh Mencari elemen terkecil dalam sebuah array Algoritme: sequential scan Orde-nya: O(n) - linear Mencari dua titik yang memiliki jarak terpendek dalam sebuah bidang (koordinat X-Y) Masalah dalam komputer grafis. Algoritme: hitung jarak dari semua pasangan titik cari jarak yang terpendek Jika jumlah titik adalah n, maka jumlah pasangan ada n * (n - 1) / 2 Orde-nya: O(n 2 ) - kuadratik Ada solusi yang O(n log n) bahkan O(n) dengan cara algoritme lain. SDA/ALG-AN/DN/V1.0/16 Fundamental Data Types: String, Math, Casting 8
Contoh (2) Tentukan apakah ada tiga titik dalam sebuah bidang yang segaris (colinier). Algoritme: periksa semua pasangan titik yang terdiri dari 3 titik. Jumlah pasangan: n * (n - 1) * (n - 2) / 6 Orde-nya: O(n 3 ) - kubik. Sangat tidak berguna untuk 10.000 titik. Ada algoritme yang kuadratik. SDA/ALG-AN/DN/V1.0/17 Studi Kasus Mengamati sebuah masalah dengan beberapa solusi. Masalah Maximum Contiguous Subsequence Sum Diberikan (angka integer negatif dimungkinkan) A 1, A 2,, A N, cari nilai maksimum dari (A i + A i+1 + + A j ). maximum contiguous subsequence sum adalah nol jika semua integer adalah negatif. (Why?) Contoh (maximum subsequences digarisbawahi) -2, 11, -4, 13, -4, 2 1, -3, 4, -2, -1, 6 SDA/ALG-AN/DN/V1.0/18 Fundamental Data Types: String, Math, Casting 9
Brute Force Algorithm (1) Algoritme: Hitung jumlah dari semua sub-sequence yang mungkin Cari nilai maksimumnya SDA/ALG-AN/DN/V1.0/19 Brute Force Algorithm (2) public static int MaxSubSum1( int [] A ) { int MaxSum = 0; for(int i = 0; i < A.length; i++) { for( int j = i; j < A.length; j++) { int ThisSum = 0; for(int k = i; k <= j; k++) ThisSum += A[ k ]; if( ThisSum > MaxSum ) MaxSum = ThisSum; return MaxSum; SDA/ALG-AN/DN/V1.0/20 Fundamental Data Types: String, Math, Casting 10
Analysis Loop of size N inside of loop of size N inside of loop of size N means O(N 3 ), or cubic algorithm. Slight over-estimate that results from some loops being of size less than N is not important. SDA/ALG-AN/DN/V1.0/21 Actual Running Time For N = 100, actual time is 0.47 seconds on a particular computer. Can use this to estimate time for larger inputs: T(N) = cn 3 T(10N) = c(10n) 3 = 1000cN 3 = 1000T(N) Inputs size increases by a factor of 10 means that running time increases by a factor of 1,000. For N = 1000, estimate an actual time of 470 seconds. (Actual was 449 seconds). For N = 10,000, estimate 449000 seconds (6 days). SDA/ALG-AN/DN/V1.0/22 Fundamental Data Types: String, Math, Casting 11
How To Improve Remove a loop; not always possible. Here it is: innermost loop is unnecessary because it throws away information. ThisSum for next j is easily obtained from old value of ThisSum: Need A i + A i+1 + + A j-1 + A j Just computed A i +A i+1 + + A j-1 What we need is what we just computed + A j SDA/ALG-AN/DN/V1.0/23 The Better Algorithm public static int MaxSubSum2( int [ ] A ) { int MaxSum = 0; for( int i = 0; i < A.length; i++ ) { int ThisSum = 0; for( int j = i; j < A.length; j++ ) { ThisSum += A[ j ]; if( ThisSum > MaxSum ) MaxSum = ThisSum; return MaxSum; SDA/ALG-AN/DN/V1.0/24 Fundamental Data Types: String, Math, Casting 12
Analysis Same logic as before: now the running time is quadratic, or O(N 2 ) As we will see, this algorithm is still usable for inputs in the tens of thousands. Recall that the cubic algorithm was not practical for this amount of input. SDA/ALG-AN/DN/V1.0/25 Actual running time For N = 100, actual time is 0.011 seconds on the same particular computer. Can use this to estimate time for larger inputs: T(N) = cn 2 T(10N) = c(10n) 2 = 100cN 2 = 100T(N) Inputs size increases by a factor of 10 means that running time increases by a factor of 100. For N = 1000, estimate a running time of 1.11 seconds. (Actual was 1.12 seconds). For N = 10,000, estimate 111 seconds (= actual). SDA/ALG-AN/DN/V1.0/26 Fundamental Data Types: String, Math, Casting 13
Recursive Algorithm Use a divide-and-conquer approach. Membagi (divide) permasalahan ke dalam bagian yang lebih kecil Menyelesaikan (conquer) masalah per bagian secara recursive Menggabung penyelesaian per bagian menjadi solusi masalah awal SDA/ALG-AN/DN/V1.0/27 Recursive Algorithm (2) The maximum subsequence either lies entirely in the first half lies entirely in the second half starts somewhere in the first half, goes to the last element in the first half, continues at the first element in the second half, ends somewhere in the second half. Compute all three possibilities, and use the maximum. First two possibilities easily computed recursively. SDA/ALG-AN/DN/V1.0/28 Fundamental Data Types: String, Math, Casting 14
Computing the Third Case Easily done with two loops; see the code For maximum sum that starts in the first half and extends to the last element in the first half, use a right-to-left scan starting at the last element in the first half. For the other maximum sum, do a left-to-right scan, starting at the first element in the first half. 4-3 5-2 -1 2 6-2 4* 0 3-2 -1 1 7* 5 SDA/ALG-AN/DN/V1.0/29 Recursion version static private int maxsumrec (int[] a, int left, int right) { int center = (left + right) / 2; if(left == right) // Base case return a[left] > 0? a[left] : 0; int maxleftsum = maxsumrec (a, left, center); int maxrightsum = maxsumrec (a, center+1, right); for(int i = center; i >= left; i--) { leftbordersum += a[i]; if(leftbordersum > maxleftbordersum) maxleftbordersum = leftbordersum;... SDA/ALG-AN/DN/V1.0/30 Fundamental Data Types: String, Math, Casting 15
Recursion Version... for(int j = center + 1; j <= right; j++) { rightbordersum += a[j]; if(rightbordersum > maxrightbordersum) maxrightbordersum = rightbordersum; return max3 (maxleftsum, maxrightsum, maxleftbordersum + maxrightbodersum); // publicly visible routine static public int maxsubsum (int [] a) { return maxsumrec (a, 0, a.length-1); SDA/ALG-AN/DN/V1.0/31 Coding Details The code is more involved Make sure you have a base case that handles zeroelement arrays. Use a public static driver with a private recursive routine. Recursion rules: Have a base case Make progress to the base case Assume it works Avoid computing the same solution twice SDA/ALG-AN/DN/V1.0/32 Fundamental Data Types: String, Math, Casting 16
Analysis Let T( N ) = the time for an algorithm to solve a problem of size N. Then T( 1 ) = 1 (1 will be the quantum time unit; remember that constants don't matter). T( N ) = 2 T( N / 2 ) + N Two recursive calls, each of size N / 2. The time to solve each recursive call is T( N / 2 ) by the above definition Case three takes O( N ) time; we use N, because we will throw out the constants eventually. SDA/ALG-AN/DN/V1.0/33 Bottom Line T(1) = 1 = 1 * 1 T(2) = 2 * T(1) + 2 = 4 = 2 * 2 T(4) = 2 * T(2) + 4 = 12 = 4 * 3 T(8) = 2 * T(4) + 8 = 32 = 8 * 4 T(16) = 2 * T(8) + 16 = 80 = 16 * 5 T(32) = 2 * T(16) + 32 = 192 = 32 * 6 T(64) = 2 * T(32) + 64 = 448 = 64 * 7 T(N) = N(1 + log N) = O(N log N) SDA/ALG-AN/DN/V1.0/34 Fundamental Data Types: String, Math, Casting 17
N log N Any recursive algorithm that solves two half-sized problems and does linear non-recursive work to combine/split these solutions will always take O( N log N ) time because the above analysis will always hold. This is a very significant improvement over quadratic. It is still not as good as O( N ), but is not that far away either. There is a linear-time algorithm for this problem; see the online code. The running time is clear, but the correctness is non-trivial. SDA/ALG-AN/DN/V1.0/35 Linear Algorithms Linear algorithm would be best. Remember: linear means O( N ). Running time is proportional to amount of input. Hard to do better for an algorithm. If input increases by a factor of ten, then so does running time. SDA/ALG-AN/DN/V1.0/36 Fundamental Data Types: String, Math, Casting 18
Idea Let A i,j be any sequence with S i,j < 0. If q > j, then A i,q is not the maximum contigous subsequence. Proof: S i,q = S i,j + S j+1,q S i,j < 0 => S i,q < S j+1,q SDA/ALG-AN/DN/V1.0/37 The Code - version 1 static public int maximumsubsequencesum3 (int a[]) { int maxsum = 0; int thissum = 0; for (int j = 0; j < a.length; j++) { thissum += a[j]; if (thissum > maxsum) { maxsum = thissum; else if (thissum < 0) { thissum = 0; return maxsum; SDA/ALG-AN/DN/V1.0/38 Fundamental Data Types: String, Math, Casting 19
The Code - version 2 static public int maximumsubsequencesum3b (int data[]) { int thissum = 0, maxsum = 0; for (int i = 0; i < data.length; i++) { thissum = Max(0, data[i] + thissum); maxsum = Max(thisSum, maxsum); return maxsum; SDA/ALG-AN/DN/V1.0/39 Running Time SDA/ALG-AN/DN/V1.0/40 Fundamental Data Types: String, Math, Casting 20
Running Time: on different machines Cubic Algorithm on Alpha 21164 at 533 Mhz using C compiler Linear Algorithm on Radio Shack TRS-80 Model III (a 1980 personal computer with a Z-80 processor running at 2.03 Mhz) using interpreted Basic n Alpha 21164A, C compiled, Cubic Algorithm TRS-80, Basic interpreted, Linear Algorithm 10 0.6 microsecs 200 milisecs 100 0.6 milisecs 2.0 secs 1,000 0.6 secs 20 secs 10,000 10 mins 3.2 mins 100,000 7 days 32 mins 1,000,000 19 yrs 5.4 hrs SDA/ALG-AN/DN/V1.0/41 Running Time: moral story Even the most clever programming tricks cannot make an ineffiecient algorithm fast. Before we waste effort attempting to optimize code, we need to optimize the algorithm. SDA/ALG-AN/DN/V1.0/42 Fundamental Data Types: String, Math, Casting 21
The Logarithm Formal Definition For any B, N > 0, log B N = K if B K = N. If (the base) B is omitted, it defaults to 2 in computer science. Examples: log 32 = 5 (because 2 5 = 32) log 1024 = 10 log 1048576 = 20 log 1 billion = about 30 The logarithm grows much more slowly than N, and slower than the square root of N. SDA/ALG-AN/DN/V1.0/43 Examples of the Logarithm BITS IN A BINARY NUMBER How many bits are required to represent N consecutive integers? REPEATED DOUBLING Starting from X = 1, how many times should X be doubled before it is at least as large as N? REPEATED HALVING Starting from X = N, if N is repeatedly halved, how many iterations must be applied to make N smaller than or equal to 1 (Halving rounds up). Answer to all of the above is log N (rounded up). SDA/ALG-AN/DN/V1.0/44 Fundamental Data Types: String, Math, Casting 22
Why log N? B bits represents 2 B integers. Thus 2 B is at least as big as N, so B is at least log N. Since B must be an integer, round up if needed. Same logic for the other examples. SDA/ALG-AN/DN/V1.0/45 Repeated Halving Principle An algorithm is O( log N ) if it takes constant time to reduce the problem size by a constant fraction (which is usually 1/2). Reason: there will be log N iterations of constant work. SDA/ALG-AN/DN/V1.0/46 Fundamental Data Types: String, Math, Casting 23
Static Searching Given an integer X and an array A, return the position of X in A or an indication that it is not present. If X occurs more than once, return any occurrence. The array A is not altered. If input array is not sorted, solution is to use a sequential search. Running times: Unsuccessful search: O( N ); every item is examined Successful search: Worst case: O( N ); every item is examined Average case: O( N ); half the items are examined Can we do better if we know the array is sorted? SDA/ALG-AN/DN/V1.0/47 Binary Search Yes! Use a binary search. Look in the middle Case 1: If X is less than the item in the middle, then look in the subarray to the left of the middle Case 2: If X is greater than the item in the middle, then look in the subarray to the right of the middle Case 3: If X is equal to the item in the middle, then we have a match Base Case: If the subarray is empty, X is not found. This is logarithmic by the repeated halving principle. The first binary search was published in 1946. The first published binary search without bugs did not appear until 1962. SDA/ALG-AN/DN/V1.0/48 Fundamental Data Types: String, Math, Casting 24
/** Performs the standard binary search using two comparisons per level. @param a the array @param x the key @exception ItemNotFound if appropriate. @return index where item is found. */ public static int binarysearch (Comparable [ ] a, Comparable x ) throws ItemNotFound { int low = 0; int high = a.length - 1; int mid; while( low <= high ) { mid = (low + high) / 2; if (a[mid].compares (x) < 0) { low = mid + 1; else if (a[mid].compares (x) > 0) { high = mid - 1; else { return mid; throw new ItemNotFound( "BinarySearch fails" ); SDA/ALG-AN/DN/V1.0/49 Binary Search Can do one comparison per iteration instead of two by changing the base case. See online code for details. Average case and worst case in revised algorithm are identical. 1 + log N comparisons (rounded down to the nearest integer) are used. Example: If N = 1,000,000, then 20 element comparisons are used. Sequential search would be 25,000 times more costly on average. SDA/ALG-AN/DN/V1.0/50 Fundamental Data Types: String, Math, Casting 25
Big-Oh Rules (1) Mathematical expression relative rates of growth DEFINITION: (Big-Oh) T(N) = O(F(N)) if there are positive constants c and N 0 such that T(N) c F(N) when N N 0. DEFINITION: (Big-Omega) T(N) = Ω(F(N)) if there are constants c and N 0 such that T(N) c F(N) when N N 0. DEFINITION: (Big-Theta) T(N) = Θ(F(N)) if and only if T(N) = O(F(N)) and T(N) = Ω(F(N)). DEFINITION: (Little-Oh) T(N) = o(f(n)) if and only if T(N) = O(F(N)) and T(N) Θ (F(N)). SDA/ALG-AN/DN/V1.0/51 Big-Oh Rules (2) Meanings of the various growth functions T(N) = O(F(N)) Growth of T(N) is growth of F(N) T(N) = Ω(F(N)) Growth of T(N) is growth of F(N) T(N) = Θ(F(N)) Growth of T(N) is = growth of F(N) T(N) = o(f(n)) Growth of T(N) is < growth of F(N) SDA/ALG-AN/DN/V1.0/52 Fundamental Data Types: String, Math, Casting 26
Limitation of Big Oh Tidak dapat digunakan untuk N yang kecil Faktor X Memory access & disk access Memory size & disk cache SDA/ALG-AN/DN/V1.0/53 Summary more data means the program takes more time Big-Oh tidak dapat digunakan untuk N yang kecil SDA/ALG-AN/DN/V1.0/54 Fundamental Data Types: String, Math, Casting 27
Exercises Urutkan fungsi berikut berdasarkan laju pertumbuhan (growth rate) N, N, N 1.5, N 2, N log N, N log log N, N log 2 N, N log (N 2 ), 2/N, 2 N, 2 N/2, 37, N 3, N 2 log N A 5 + B 5 + C 5 + D 5 + E 5 = F 5 0 < A B C D E F 75 hanya memiliki satu solusi. berapa? SDA/ALG-AN/DN/V1.0/55 Assignment Critical Mass Secara umum dapat dilihat di http://pala.acad.cs.ui.ac.id/contest/ OnlineJudge/v5/580.html Lengkapnya akan diposting di forum.iki.struktur Due date: 2 Maret 2001 SDA/ALG-AN/DN/V1.0/56 Fundamental Data Types: String, Math, Casting 28
Further Reading Chapter 5 http://pala.acad.cs.ui.ac.id/contest/ mirror/evo.apm.tuwien.ac.at/ AlgDesignManual/INDEX.HTM http://pala.acad.cs.ui.ac.id/contest/ OnlineJudge/v1/108.html SDA/ALG-AN/DN/V1.0/57 What s Next Recursion (Chapter 7) SDA/ALG-AN/DN/V1.0/58 Fundamental Data Types: String, Math, Casting 29