Transcript Sorting - Universidade Federal de Campina Grande
Sorting
Introduction
• The objective is to take an unordered set of comparable data items and arrange them in order.
• We will usually sort the data into ascending order — sorting into descending order is very similar.
• Data can be sorted in various ADTs, such as arrays and trees; in fact, we have already seen how a binary search tree can be used to sort data.
• There are two main sorting themes: address-based sorting and comparison-based sorting. We will focus on the latter.
2 CSM1220 - Sorting
The family of sorting methods
Main sorting themes Comparison-based sorting Transposition sorting BubbleSort Insert and keep sorted Insertion sort
CSM1220 - Sorting
Tree sort Diminishing increment sorting ShellSort Proxmap Sort Address -based sorting RadixSort Priority queue sorting Selection sort Heap sort Divide and conquer QuickSort MergeSort
3
Lecture schedule
• • An
overview
of sorting algorithms • We
will not study address-based sorting
in detail because they are quite restricted in their application – One slide on Proxmap/Radix sort
Examples
from each type
of comparison-based sorting
algorithm – Bubble sort [transposition sorting] – Insertion sort (already seen, see Tree notes) [insert and keep sorted] – Selection sort [priority queue sorting] – Shell sort [diminishing increment sorting] – Merge sort and Quick sort [divide and conquer sorting] CSM1220 - Sorting 4
Bubble sort
• A pretty dreadful type of sort!
end of one inner loop • However, the code is small: for (int i=arr.length; i>0; i--) { for (int j=1; j arr[j]) { temp = arr[j-1]; arr[j-1] = arr[j]; arr[j] = temp; } } } 5 ‘bubbled’ to the correct position remaining elements put in place 5 3 3 3 2 3 5 2 2 3 2 2 5 4 4 5 4 4 4 5 CSM1220 - Sorting 5
Insertion sort
Tree Insertion Sort
• This is inserting into a normal tree structure: i.e. data are put into the correct position when they are inserted.
• We’ve already seen this happening with a tree, requiring a find and an insert.
• We know the time complexity for one insert is O(
logN
) + O(1) = O(
logN
); therefore to insert
N
items will have a complexity of O(
NlogN
).
Array Insertion Sort
• The array must be sorted; insertion requires a find + insert.
• The insertion time complexity is O(
logN
) + O(
N
) = O(
N
) for one item, therefore to insert
N
items will have a complexity of O(
N 2
).
CSM1220 - Sorting 6
SelectionSort
• SelectionSort uses an
array
implementation of the Priority Queue ADT (not the heap implementation we will study) • The elements are inserted into the array as they arrive and then extracted in descending order into another array • The major disadvantage is the performance overhead of finding the largest element at each step; we have to traverse over the entire array to find it 13 2 15 4 15 13 4 2 7 CSM1220 - Sorting
SelectionSort (2)
• In practice, we use the same array for the Priority Queue and the output results • General algorithm: 1. Initialise PQLast to the last index of the Priority Queue 2. Search from the start of the array to PQLast for the largest element: call its position front 3. Swap element indexed by front with element indexed by PQLast 4. Decrement PQLast by one 5. Repeat steps 2 – 4 CSM1220 - Sorting 8
SelectionSort (3)
CSM1220 - Sorting 13 2 15 4 12 9 13 2 front 9 4 PQLast 12 15 front 12 2 9 4 PQLast 13 15 front
etc.
PQLast 9
Time complexity of SelectionSort
• The main operations being performed in the algorithm are: 1. Comparisons to find the largest element in the Priority Queue subarray (to find front index): there are ‘
n
+ (
n
-1) + … + 2 + 1’ comparisons, i.e.,
n/
2 (
n
+1) = (
n
2 +
n
) / 2 so this is an O(
n
2 ) operation 2. Swapping elements between front and PQLast:
n
-1 exchanges are performed so this is an O(
n
) operation • The dominant operation (time-wise) gives the overall time complexity, i.e., O(
n
2 ) • Although this is an O(
n
2 ) algorithm, its advantage over O(
n
log
n
) sorts is its simplicity 10 CSM1220 - Sorting
Time complexity of SelectionSort (2)
• For very small sets of data, SelectionSort may actually be more efficient than O(
n
log
n
) algorithms • This is because many of the more complex sorts have a relatively high level of overhead associated with them, e.g., recursion is expensive compared with simple loop iteration • This overhead might outweigh the gains provided by a more complex algorithm where a small number of data elements is being sorted • SelectionSort does better than BubbleSort as fewer swaps are required, although the same number of comparison operations are performed (each swap puts an element in its correct place) 11 CSM1220 - Sorting
Shell sort
[diminishing increment sorting]
• Named after D.L. Shell! But it is also rather like shrinking shells of sorting, for Insertion Sort.
• Shell sort aims to reduce the work done by insertion sort (i.e. scanning a list and inserting into the right position).
• Do the following: – Begin by looking at the lists of elements
x
1 those elements by insertion sort elements apart and sort – Reduce the number
x
1 these two steps until
x
2 to
x
2 = 1.
so that
x
1 is not a multiple of
x
2 and repeat • It can be shown that this approach will sort a list with a time complexity of O(N 1.25
).
CSM1220 - Sorting 12
Shell Sort Illustration
45 23 4 9 6 7 91 8 12 24 • Consider sorting the following list by Shell sort: GAP=3 GAP=2 GAP=1 9 6 4 24 8 7 45 91 23 12 9 6 4 24 8 7 45 23 12 91 4 6 8 7 9 23 12 24 45 91 9 23 12 24 4 6 8 7 45 91 4 6 7 8 9 12 23 24 45 91 CSM1220 - Sorting 13
Comparing O(N
2
), O(N
1.25
) and O(N)
O(N) O(N^1.25) 1 1 2 2.378414
3 3.948222
4 5.656854
5 7.476744
6 9.390507
7 11.38604
8 13.45434
9 15.58846
10 17.78279
11 20.03276
12 22.33452
13 24.68478
14 27.08071
15 29.51985
16 32 17 34.51923
18 37.07581
19 39.66815
20 42.29485
21 44.9546
22 47.64621
23 50.36859
24 53.12073
25 55.9017
O(N^2) 1 4 9 16 25 36 49 64 81 100 121 144 169 196 225 256 289 324 361 400 441 484 529 576 625 300 200 100 0 700 600 500 400 O(N) O(N^1.25) O(N^2) Size of input (N) 14 CSM1220 - Sorting
How do you choose the gap size?
• The idea of the decreasing gap size is that the list becomes more and more sorted each time the gap size is reduced, therefore you don’t (for example) want to have a gap size of 4 followed by a gap size of 2 because you’ll be sorting half the numbers a second time. • There is no formal proof of a good initial gap size, but about a 10 th the size of N is a reasonable start. • Try to use prime numbers as your gap size, or odd numbers if you can not readily get a list of primes (though note gaps of 9, 7, 5, 3, 1 will be doing less work when gap=3).
15 CSM1220 - Sorting
MergeSort
Divide and conquer sorting
QuickSort 16 CSM1220 - Sorting
Divide ...
5 1 4 2 10 3 9 15 5 1 4 2 10 3 9 15 5 1 4 2 10 3 9 15 5 CSM1220 - Sorting 1 4 2 10 3 9 15 17
and conquer
1 2 3 4 5 9 10 15 1 2 4 5 3 9 10 15 1 5 2 4 3 10 9 15 5 CSM1220 - Sorting 1 4 2 10 3 9 15 18
MergeSort
[divide and conquer sorting]
• For MergeSort an initial array is repeatedly divided into halves (usually each is a separate array), until arrays of just one or zero elements remain • At each level of recombination, two sorted arrays are merged into one • This is done by copying the smaller of the two elements from the sorted arrays into the new array, and then moving along the arrays 1 13 24 26 2 15 27 38 Aptr CSM1220 - Sorting Bptr Cptr 19
Merging
1 13 24 26 2 15 27 38 1 1 Aptr 13 24 26 Bptr 2 15 27 38 1 Cptr 2 1 Aptr 13 24 26 2 Bptr 15 27 38 1 2 Cptr 13 Aptr Bptr Cptr
etc.
CSM1220 - Sorting 20
Merge Sort In Pseudocode
• Basic idea in pseudocode:
method
mergeSort(array)
is
:
if
len(array) == 0
return
array
or
len(array) == 1
then
:
// terminating case
else
: end = len(array) center = end / 2 left = mergeSort( array[0:center] )
// recursive call
right = mergeSort( array[center:end] )
// recursive call
return
merge(left,right)
// method that merges lists
end if end method
CSM1220 - Sorting 21
Analysis of MergeSort
• Let the time to carry out a MergeSort on
n
elements be
T
(
n
) • Assume that
n
is a power of 2, so that we always split into equal halves (the analysis can be refined for the general case) • For
n
=1, the time is constant, so we can take
T
(1) = 1 • Otherwise, the time
T
(
n
) is the time to do two MergeSorts on
n
/2 elements, plus the time to merge, which is linear • So,
T
(
n
) = 2
T
(
n
/2) +
n
• Divide through by
n
to get
T
(
n
)/
n
=
T
(
n
/2)/(
n
/2) + 1 • Replacing
n
by
n
/2 gives,
T
(
n
/2)/(
n
/2) =
T
(
n
/4)/(
n
/4) + 1 • And again gives,
T
(
n
/4)/(
n
/4) =
T
(
n
/8)/(
n
/8) + 1 CSM1220 - Sorting 22
Analysis of MergeSort (2)
• We continue until we end up with
T
(2)/2 =
T
(1)/1 + 1 • Since
n
is divided by 2 at each step, we have log 2
n
steps • Now, substituting the last equation in the previous one, and working back up to the top gives
T
(
n
)/
n
=
T
(1)/1 + log 2
n
• That is,
T
(
n
)/
n
= log 2
n
• So
T
(
n
) =
n
log 2
n
+
n
+ 1 = O(
n
log • Although this is an O(
n
log
n n
) ) algorithm, it is hardly ever used for main memory sorts because it requires linear extra memory 23 CSM1220 - Sorting
QuickSort
[divide and conquer sorting]
• As its name implies, QuickSort is the fastest known sorting algorithm
in practice
(address-based sorts can be faster) • It was devised by C.A.R. Hoare in 1962 • Its average running time is O(
n
log
n
) and it is very fast • It has worst-case performance of O(
n
2 ) but this can be made very unlikely with little effort • The idea is as follows: 1. If the number of elements to be sorted is 0 or 1, then return 2. Pick any element,
v
(this is called the
pivot
) 3. Partition the other elements into two disjoint sets,
S
1 and
S
2 of elements >
v
4. Return QuickSort (
S
1 ) followed by
v
of elements followed by QuickSort (
S
2 )
v
, CSM1220 - Sorting 24
QuickSort example
5 1 4 2 10 3 9 15 12 Pick the middle element as the pivot, i.e., 10 Partition into the two subsets below 5 1 4 2 3 9 15 12 Sort the subsets 1 2 3 4 5 Recombine with the pivot 1 2 3 4 5 9 9 10 12 12 15 15 CSM1220 - Sorting 25
QuickSort Example (full)
Unsorted List 5 1 4 2 10 3 9 15 12 Partition1 Partition2 Partition3 5 1 4 2 3 9 10 15 12 1 1 2 3 2 3 4 5 9 5 9 12 15 Sorted List 1 2 3 4 5 9 10 12 15 CSM1220 - Sorting 26
QuickSort Pseudocode
• Basic idea in pseudocode
(compare with mergeSort)
method
quickSort(array)
is
:
if
len(array) == 0
return
array
or
len(array) == 1
then
:
// terminating case
else
: pivotIndx = int( random() * len(array) ) pivot = array(pivotIndx)
// just choose a random pivot
(left, right) = partition(array, pivot)
// get left and right
return
quickSort(left) + pivot + quickSort(right)
end if end method
CSM1220 - Sorting 27
Partitioning example
5 11 4 25 10 3 9 15 12 Pick the middle element as the pivot, i.e., 10 Move the pivot out of the way by swapping it with the first element 10 11 4 25 5 3 9 15 12 swapPos Step along the array, swapping small elements into swapPos 10 4 11 25 5 3 9 15 12 swapPos CSM1220 - Sorting 28
CSM1220 - Sorting
Partitioning example (2)
10 4 5 25 11 3 9 15 12 10 4 5 swapPos 3 11 25 9 15 12 swapPos 10 4 5 3 9 25 11 15 12 swapPos 29
Pseudo code for partitioning
pivotPos = middle of array a; swap a[pivotPos] with a[first]; // Move the pivot out of the way swapPos = first + 1; for each element in the array from swapPos to last do: // If the current element is smaller than pivot we // move it towards start of array if (a[currentElement] < a[first]): swap a[swapPos] with a[currentElement]; increment swapPos by 1; // Now move the pivot back to its rightful place swap a[first] with a[swapPos-1]; return swapPos-1; // Pivot position CSM1220 - Sorting 30
Analysis of QuickSort
• We assume a random choice of pivot • Let the time to carry out a QuickSort on
n
elements be
T
(
n
) • We have
T
(0) =
T
(1) = 1 • The running time of QuickSort is the running time of the partitioning (linear in n) plus the running time of the two recursive calls of QuickSort • Let
i
be the number of elements in the left partition, then
T
(
n
) =
T
(
i
) +
T
(
n–i
–1) +
cn
(for some constant
c
) CSM1220 - Sorting 31
Worst-case analysis
• If the pivot is always the smallest element, then
i
= 0 always • We ignore the term
T
(0) = 1, so the recurrence relation is
T
(
n
) =
T
(
n–
1) +
cn
• So,
T
(
n–
1) =
T
(
n–
2) +
c
(
n–
1) and so on until we get
T
(2) =
T
(1) +
c
(2) • Substituting back up gives
T
(
n
) =
T
(1) +
c
(
n
+ … + 2) = O(
n
2 ) • Notice that this case happens if we always take the pivot to be the first element in the array and the array is already sorted • So, in this extreme case, QuickSort takes O(
n
2 ) time to do absolutely nothing!
CSM1220 - Sorting 32
Best-case analysis
• In the best case, the pivot is in the middle • To simplify the equations, we assume that the two subarrays are each exactly half the length of the original (a slight overestimate which is acceptable for big-Oh calculations) • So, we get
T
(
n
) = 2
T
(
n/2
) +
cn
• This is very similar to the formula for MergeSort, and a similar analysis leads to
T
(
n
) =
cn
log 2
n
+
n
= O(
n
log
n
) 33 CSM1220 - Sorting
Average-case analysis
• We assume that each of the sizes of the left partition are equally likely, and hence have probability 1/
n
• With this assumption, the average value of
T
(
i
), and hence also of
T
(
n–i
–1), is (
T
(0) +
T
(1) + … +
T
(
n–
1))/
n
• Hence, our recurrence relation becomes
T
(
n
) = 2(
T
(0) +
T
(1) + … +
T
(
n–
1))/
n
+
cn
• Multiplying by
n
gives
nT
(
n
) = 2(
T
(0) +
T
(1) + … +
T
(
n–
1)) +
cn
2 • Replacing
n
by
n–
1 gives (
n–
1)
T
(
n–
1) = 2(
T
(0) +
T
(1) + … +
T
(
n–
2)) +
c
(
n–
1) 2 • Subtracting the last equation from the previous one gives
nT
(
n
) – (
n–
1)
T
(
n–
1) = 2
T
(
n–
1) + 2
cn
– c CSM1220 - Sorting 34
Average-case analysis (2)
• Rearranging, and dropping the insignificant
c
gives
nT
(
n
) = (
n+
1)
T
(
n–
1) + 2
cn
• Divide through by
n
(
n+
1) to get
T
(
n
)/(
n+
1) =
T
(
n–
1)/
n
+ 2
c
/(
n+
1) on the end, • Hence,
T
(
n–
1)/
n
=
T
(
n–
2)/(
n–
1) + 2
c
/
n T
(2)/3 =
T
(1)/2 + 2
c
/3 and so on down to • Substituting back up gives
T
(
n
)/(
n+
1) =
T
(1)/2 + 2
c
(1/3 + 1/4 + … + 1/(
n+
1)) • The sum in brackets is about log
e
(
n
+1) + – 3/2, where Euler’s constant, which is approximately 0.577
• So,
T
(
n
)/(
n+
1) = O(log
n
) and
T
(
n
) = O(
n
log
n
) is CSM1220 - Sorting 35
Some observations about QuickSort
• We have seen that a consistently poor choice of pivot can lead to O(
n
2 ) time performance • A good strategy is to pick the middle value of the left, centre, and right elements • For small arrays, with
n
less than (say) 20, QuickSort does not perform as well as simpler sorts such as SelectionSort • Because QuickSort is recursive, these small cases will occur frequently • A common solution is to stop the recursion at
n
middle of three elements when
n
is one or two = 10, say, and use a different, non-recursive sort • This also avoids nasty special cases, e.g., trying to take the CSM1220 - Sorting 36
Address-Based Sorting
• Proxmap uses techniques similar to hashing to assign an element to its correctly sorted position in a container such as an array using a hash function • The algorithms are generally complex, and very often only suitable for certain kinds of data • You need to find a suitable hashing function that will distribute data fairly evenly • Clashes are dealt with using a comparison based sort, so the more clashes there are the further the time complexity moves away from O(n)
(see notes view for more details)
37 CSM1220 - Sorting
Radix / Bucket / Bin sort
• Radix Sort – In its favour, it is an
O(n)
investigated. sort. That makes it the fastest sort we have – However it requires at least
2n
space in which to operate, and is in principle exponential in its space complexity.. – A radix sort makes one pass through the data for each
atom
in the key. If the key is three character uppercase alphabetic strings, such as
ABC
, then
A
is an atom. In this case, there would be three passes through the data. – The first pass sorts on the low order element of the key (the
C
in our example). The second pass sorts on the next atom, in order of importance (the
B
in our example). Each pass progresses toward the high order atom of the key. – In each such pass, the elements of the array are distributed into
k
buckets. For our alphabetic key example,
A
's go into the 'A' bucket,
B
's into the 'B' bucket, and so on. These distribution buckets are then gathered, in order, and placed back in the original array. The next pass is then executed. When a pass has been made on the high order atom in the key, the array will be sorted. 38 CSM1220 - Sorting
123 234 345 134 245 356 156 267 378 121 232 343 CSM1220 - Sorting
Radix Sort Example
col1 col2 col3 12
1 1
23 134 156 121 1
2
3 121 1
3
4 1
5
6 12
3
Bins
0
,
2
,
4
-
9
Are empty
2
34 245 267 232 Bins
0
-
1
,
4
,
6
-
9
Are empty 2
3
4 232 134 156 23
2 3
45 356 378 343 Bins
0
,
4
-
9
are empty 2
4
5 2
6
7 Bins
0
-
2
,
5
,
7
-
9
Are empty 23
4 .
.
.
Bins
0
-
1
,
3
,
5
-
9
Are empty 39