Scribd
Upload
34 views

Sorting

The document discusses sorting algorithms. It introduces selection sort and quicksort as examples of sorting algorithms. Selection sort works by iterating through a list and selecting the smallest element, adding it to a sorted list. Quicksort works by selecting a pivot element and partitioning the list into elements less than and greater than the pivot, then recursively sorting the sublists. The document analyzes the time complexity of these algorithms and proves a lower bound of Ω(N log N) comparisons for any sorting algorithm.

Uploaded by

Siva Kumar
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
34 views

Sorting

The document discusses sorting algorithms. It introduces selection sort and quicksort as examples of sorting algorithms. Selection sort works by iterating through a list and selecting the smallest element, adding it to a sorted list. Quicksort works by selecting a pivot element and partitioning the list into elements less than and greater than the pivot, then recursively sorting the sublists. The document analyzes the time complexity of these algorithms and proves a lower bound of Ω(N log N) comparisons for any sorting algorithm.

Uploaded by

Siva Kumar
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 15

Lecture 11: Sorting Algorithms

CS105: Great Insights in Computer Science Michael L. Littman, Fall 2006

Sorting Algorithms
Another name for the lecture is Google II. Sorting is a great topic in CS: relatively simple extremely important illustrates lots of different algorithms and analysis techniques Theres more than one way to skin a cat.

Google...
Last time, I said Google does its thing in a couple of very signicant steps: I. Collect pages from the web (graph search). II. Index them. III.Respond to queries.

Three Pages of Words


3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 html head title monica babes rutgers title head frameset rows frame name banner scrolling no noresize target contents src header htm frameset cols frame name contents target main src left htm frame name main src right htm frameset noframes body 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 p this page uses frames but your browser doesn t support them p body noframes frameset html doctype html public w c dtd html nal en html head title index of mlittman courses cs python title head body h index 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 of mlittman courses cs python h ul li a href mlittman courses cs parent directory a li a href add py add py a li a href differences old py differences old py a li a href differences py differences 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 py a li a href exception py exception py a li a href ml py ml py a li a href ml b py ml b py a li a href nim nim a li a href nim py nim 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 py a li a href nim py nim py a li a href nim py nim py a li a href python walkthru pdf python walkthru pdf a li a href socks py socks py a ul address apache server 4 4 4 4 4 4 4 4 4 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 at www cs rutgers edu port address body html header title google test unlinked page title header html h how does google nd pages h this page has little purpose other than to include the word googleblackout this is a 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 word sort of that i concocted on january and veried that it was unknown to google your search b googleblackout b did not match any documents p there is a primary page http www cs rutgers edu mlittman courses cs 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 googletest html that is explicitly linked from the course homepage it has its own special term which consists of concatenating google and discovery p visit the primary page for the search history p this page is part of a href 9 9 9 9 9 9 9 9 9 index html great insights in computer science a html

What Can We Do?


All the information is there, and we can sift through it. But, its slow and error-prone to skim through every page every time we want to nd something. If there are N words (total) on the web pages, how long would it take to sift through them each time? (Use big O notation.) How can we organize the data to simplify?

Sort, Remove Duplicates


4 9 4 4 9 9 4 4 4 9 3 3 3 4 3 3 4 3 9 9 9 9 3 a a add address and any apache at b b babes banner body body browser but c cols computer concatenating concocted consists contents 9 4 9 4 9 9 4 4 9 4 9 9 3 4 4 9 4 4 9 4 9 9 3 course courses courses cs cs did differences directory discovery doctype documents does doesn dtd edu edu en exception explicitly nal nd for frame 3 3 9 9 9 9 9 4 9 9 3 4 3 9 9 9 9 4 9 3 3 4 9 frames 9 frameset 9 from 9 google 9 googleblackout4 googletest 9 great 9 h 9 h 9 has 9 head 9 head 3 header 4 header 9 history 9 homepage 3 how 9 href 4 href 4 htm 9 html 3 html 3 html 4 http i in include index index insights is it its january left li linked little main match ml mlittman mlittman monica name nim 3 3 3 9 4 9 4 9 9 9 3 9 3 9 9 4 9 4 4 9 4 9 4 no noframes noresize not of of old on other own p p page page pages parent part pdf port primary public purpose py 4 3 3 3 4 9 9 3 9 4 4 9 9 3 3 3 3 9 9 9 9 9 3 python right rows rutgers rutgers rutgers science scrolling search server socks sort special src support t target term test than that the them 9 3 9 3 4 9 9 4 9 9 3 9 9 4 4 9 9 9 4 9 3 9 there this this title title title to ul unknown unlinked uses veried visit w walkthru was which word www www your your

Selection Sort
Idea is quite simple. We go through the list one item at a time. We keep track of the smallest item weve found. When were through the list, we pull the smallest item out and add it to a list of sorted items. We repeat until all the items have been removed.

Code
def Selection(l): sorted = [] while len(l) > 0: (smallest,rest) = ndSmallest(l) sorted = sorted + [smallest] l = rest return sorted def ndSmallest(l): smallest = l[0] rest = [] for i in range(1,len(l)): if l[i] < smallest: rest = rest + [smallest] smallest = l[i] else: rest = rest + [l[i]] return (smallest, rest)

Selection Sort Analysis


How many comparisons does Selection Sort do in the worst case? Assume the list is length N. Hint: What song is it like? You can use big O notation. Does it matter whether the list is sorted or not?

Other Sorting Approaches


How else can you imagine sorting? Fewer comparisons than O(N2)? bubblesort counting sort insertion sort Shell sort

Guess Who?
Each player picks a character. Players take turns asking each other yes/ no questions. First player to uniquely identify the other players character wins!

Mindreader: Set Cards

Cross-Hatched?

Squiggle?

Insight
Each question splits the remaining set of possibilities into two subsets (yes and no). We want to pick a question so that the larger of the two subsets is as small as possible. Half! How many questions? n=1, questions = 0 n=2, questions = 1 n=4, questions = 2 n=8, questions = 3 n=16, questions = 4 n, questions = lg n.

Binary Search
Lets say we have a sorted list of n items. How many comparisons do we need to make to nd where a new item belongs in the list? Can start at the bottom and compare until the new item is bigger. Maximum number of comparisons? One for each position: n. We can ask better questions: bigger than the halfway mark? That gets us: lg (n+1)!

abcdefg

Binary Search Sort


Using O(lg N) comparisons, can nd where to insert the next item. Since we insert N items, comparisons is O(N lg N) in total. Cant quite implement it that way, though: Once we nd the spot, O(N) to stick it in. However, other algorithms are really O(N lg N). Hillis mentions Quick Sort and Merge Sort.

Quicksort
quicksort: Another sorting algorithm. Idea: Break the list of n+1 elements into the median and two lists of n/2. The two lists are those smaller than the median and those larger than the median. Sort the two lists separately. Glue them together: All n are sorted.

Quicksort Example
Original list: [56, 80, 66, 64, 37, 36, 91, 48, 17, 20, 86, 89, 41, 1, 96, 12, 74] Median is 56; smaller: [37, 36, 48, 17, 20, 41, 1, 12] bigger: [80, 66, 64, 91, 86, 89, 96, 74] Sort each; smaller: [1, 12, 17, 20, 36, 37, 41, 48] bigger: [64, 66, 74, 80, 86, 89, 91, 96] Glue: [1, 12, 17, 20, 36, 37, 41, 48, 56, 64, 66, 74, 80, 86, 89, 91, 96]

But...
If we could nd the median, the whole sorting process would be pretty easy. Sufcient to split anywhere in the middle half at least half the time: Still O(n log n). Pick a random list element. 25% of the time, it will be in the 1st quarter of the sorted list, 25% of the time in the last quarter, and 50% in the middle half.

Quicksorts Flow
Pick an item, any item (the pivot). Partition the list as to less (left) or greater than (right) pivot. Sort the two halves (recursively).

Code
def Quicksort(l): if len(l) <= 1: return l pivot = l[randint(0,len(l)-1)] (left,equal,right) = partition(l,pivot) return Quicksort(left) + equal + Quicksort(right) def partition(l,pivot): left = [] right = [] equal = [] for item in l: if item < pivot: left = left + [item] if item > pivot: right = right + [item] if item == pivot: equal = equal + [item] return (left,equal,right)

Merge Sort
View all the items as separate sorted lists. Pick the two shortest lists and combine them into a single sorted list: Compare the rst items. Move smaller one to end of the combined list. Repeat until one list is empty. Repeat until only a single list is left.

Merge Sort Analysis


To merge two lists of length N requires at most 2N comparisons. If there are N items in L lists of length N/L, after one merging pass, N comparisons, L /2 lists of length 2N/L. Length doubles each time. Initially, L = N lists of length 1 each. After lg N merging passes, 1 list of length N. Total comparisons: O(N lg N).

Lower Bound
Weve shown that we can sort in O(N lg N) comparisons. What if someone comes along and does it better? We need to protect ourselves and prove a lower bound: that is, to show that nothing less than N lg N will sufce. Lets return to Guess Who?.

Sorting Lower Bound


If we are asking yes/ no questions to uniquely identify one item out of n, how many questions do we need in the worst case? Might be as many as lg n, since each question cannot exclude more than half. Sorting N elements identies the correct ordering using just yes/no questions.

Counting Orderings
How many ways to order N elements? 1: 1 2: 2 3: 6 = 3 x 2 4: 24 = 4 x 3 x 2 5: 120 = 5 x 4 x 3 x 2 N: N! = N x (N-1) x (N-2) x ... x 2 x 1 Known as the factorial function. Thus, sorting must nd the unique sorted ordering from a set of N! possibilities using just yes/no questions.

A Little Math
N! = 1 x 2 x 3 x ... x N/2 x (N/2 + 1) x ... x N > (N/2 + 1) x ... x N > N/2 x ... x N/2 = N/2 N/2 Number of comparisons to sort N items > # of yes/no questions to pick one out of N! > # of yes/no questions to pick one out of N/2 > lg N/2 N/2 = N/2 lg N/2 or, essentially N lg N. O(N lg N) wins!

N/2

Web Search, Again


Weve seen two of the major steps needed to implement a web search engine: gather up pages using graph search index the words using sorting In a later lecture, well talk about the last step: using more than one computer to respond quickly to millions of queries a day.

You might also like