Sorting
by Tammy Bailey
Algorithm analysis
• Determine the amount of resources an algorithm
requires to run
– computation time, space in memory
• Running time of an algorithm is the number of basic
operations performed
– additions, multiplications, comparisons
– usually grows with the size of the input
– faster to add 2 numbers than to add 2,000,000!
Running time
• Worst-case running time
– upper bound on the running time
– guarantee the algorithm will never take longer to run
• Average-case running time
– time it takes the algorithm to run on average
(expected value)
• Best-case running time
– lower bound on the running time
– guarantee the algorithm will not run faster
Comparisons in insertion sort
• Worst case
– element k requires (k-1) comparisons
– total number of comparisons:
0+1+2+ … + (n-1) = ½ (n)(n-1)
= ½ (n2-n)
• Best case
– elements 2 through n each require one comparison
– total number of comparisons:
1+1+1+ … + 1 = n-1
(n-1) times
Running time of insertion sort
• Best case running time is linear
• Worst case running time is quadratic
• Average case running time is also quadratic
– on average element k requires (k-1)/2 comparisons
– total number of comparisons:
½ (0+1+2+ … + n-1) = ¼ (n)(n-1)
= ¼ (n2-n)
Mergesort
27 10 12 20
divide
27 10 12 20
divide divide
27 10 12 20
merge merge
10 27 12 20
merge
10 12 20 27
Merging two sorted lists
first list second list result of merge
10 27 12 20 10
10 27 12 20 10 12
10 27 12 20 10 12 20
10 27 12 20 10 12 20 27
Comparisons in merging
• Merging two sorted lists of size m requires at least m and
at most 2m-1 comparisons
– m comparisons if all elements in one list are smaller
than all elements in the second list
– 2m-1 comparisons if the smallest element alternates
between lists
Logarithm
• Power to which any other number a must be raised to
produce n
– a is called the base of the logarithm
• Frequently used logarithms have special symbols
– lg n = log2 n logarithm base 2
– ln n = loge n natural logarithm (base e)
– log n = log10 n common logarithm (base
10)
• If we assume n is a power of 2, then the number of times
we can recursively divide n numbers in half is lg n
Comparisons at each merge
#lists #elements #merges #comparisons #comparisons
in each list per merge total
n 1 n/2 1 n/2
n/2 2 n/4 3 3n/4
n/4 4 n/8 7 7n/8
2 n/2 1 n-1 n-1
Comparisons in mergesort
• Total number of comparisons is the sum of the number
of comparisons made at each merge
– at most n comparisons at each merge
– the number of times we can recursively divide n
numbers in half is lg n, so there are lg n merges
– there are at most n lg n comparisons total
Comparison of sorting algorithms
• Best, worst and average-case running time of mergesort
is (n lg n)
• Compare to average case behavior of insertion sort:
n Insertion sort Mergesort
10 25 33
100 2500 664
1000 250000 9965
10000 25000000 132877
100000 2500000000 1660960
Quicksort
• Most commonly used sorting algorithm
• One of the fastest sorts in practice
• Best and average-case running time is O(n lg n)
• Worst-case running time is quadratic
• Runs very fast on most computers when implemented
correctly
Searching
Searching
• Determine the location or existence of an element in a
collection of elements of the same type
• Easier to search large collections when the elements are
already sorted
– finding a phone number in the phone book
– looking up a word in the dictionary
• What if the elements are not sorted?
Sequential search
• Given a collection of n unsorted elements, compare each
element in sequence
• Worst-case: Unsuccessful search
– search element is not in input
– make n comparisons
– search time is linear
• Average-case:
– expect to search ½ the elements
– make n/2 comparisons
– search time is linear
Searching sorted input
• If the input is already sorted, we can search more
efficiently than linear time
• Example: “Higher-Lower”
– think of a number between 1 and 1000
– have someone try to guess the number
– if they are wrong, you tell them if the number is higher
than their guess or lower
• Strategy?
• How many guesses should we expect to make?
Best Strategy
• Always pick the number in the middle of the range
• Why?
– you eliminate half of the possibilities with each guess
• We should expect to make at most
lg1000 10 guesses
• Binary search
– search n sorted inputs in logarithmic time
Binary search
• Search for 9 in a list of 16 elements
1 3 4 5 5 7 9 10 11 13 14 18 20 22 23 30
1 3 4 5 5 7 9 10
5 7 9 10
9 10
9
Sequential vs. binary search
• Average-case running time of sequential search is linear
• Average-case running time of binary search is logarithmic
• Number of comparisons:
n sequential binary
search search
2 1 1
16 8 4
256 128 8
4096 2048 12
65536 32768 16
