Transcript CS342 Data Structures - William Paterson University

CS342 Data Structures
Complexity Analysis
Hu:
asymptotic:
Complexity Analysis: topics
•
•
•
•
•
•
•
•
•
Introduction
Simple examples
The Big-O notation
Properties of Big-O notation
Ω and Θ notations
More examples of complexity analysis
Asymptotic complexity
The best, average, and worst cases
Amortized complexity
Definition of a good program
• Correctness: does it work?
• Ease of use: user friendly?
• Efficiency: runs fast? consumes a lot of
memory?
• Others
– easy to maintain
–…
Why it is important to study
complexity of algorithm
• It addresses the efficiency (time and space)
problem.Complexity analysis allows one to
estimate
– the time needed to execute an algorithm (time
complexity)
– the amount of memory needed to execute the
algorithm (space complexity).
Space complexity
• Instruction space: space needed to store compiled
code.
• Data space: space needed to store all constants,
variables, dynamically allocated memory,
environmental stack space (used to save
information needed to resume execution of
partially completed functions).
Although memory is inexpensive today, it can be
important for portable devices (where size, power
consumption, are important).
Time or computational complexity
• It is not objective to simply measure the total execution
time of an algorithm because the the execution time
depends on many factors such as the processor
architecture, the efficiency of the compiler, and the
semiconductor technology used to make the processor.
• One more meaningful approach is to identify one or more
key or primary operations and determine the number of
times these operations are performed. Key operations are
those that take much more time to execute than other types
of operations.
The time needed to execute an algorithm increases as the amount of data,
n, as n increases, so does the execution time. The big-O notation
provides an (asymptotic) estimate (upper bound) of the execution time
when n becomes large.
Example on key instruction count: locating the
minimum element in any array of n elements
• Assuming the comparison operation is the
key operation (i.e., assignment operation is
not considered a key operation here).
then: T(n) = n – 1
A more involved example: the selection sort
of an n-element array a into ascending order
• Again, assuming that comparison is the key
operation or instruction.
– The algorithm requires (n –1) passes; each pass locates
the minimum value in the unsorted sub-list and the
value is swapped with a[i], where i = 0 for pass 1, 1 for
pass 2, and so on. In pass 1, there are (n - 1)
comparison operations. The # comparison decreases by
1 in the following pass, which lead to:
• T(n) = (n – 1) + (n – 2) + … + 3 + 2 + 1
= n(n - 1) / 2
= n2 / 2 – n / 2
The best, worst, and average case
• Using linear search as an example:
searching for an element in an n-element
array.
– best case:
T(n) = 1
– worst case: T(n) = n
– average case: T(n) = n / 2
• The complexity analysis should take into
account of the best, worst and average
cases, although the worst case is the most
important.
The big-O notation
• Big-O notation can be obtained from T(n)
by taking the dominant term from T(n);
dominant term is the term that increases
fastest as the value of n increases.
• Example
– T(n) = n – 1, n is the dominant term (why?), therefore,
O(n) = n, where O(n) reads big-O of n.
– T(n) = n2 / 2 – n / 2, (n2 / 2) is the dominant term
(why?), after dropping the constant 2, we obtain O(n2).
We say the selection sort has a running time of O(n2)
Orders of magnitude of Big-O functions
• In reality, a small set of big-O functions defines the running
time of most algorithms.
– constant time: O(1), e.g., access first element of an array, append
an element to the end of a list, etc.
– Linear: O(n), linear search
– Quadratic: O(n2), selection sort
– Cubic: O(n3); matrix multiplication; slow
– Logarithmic: O(log2n), binary search; very fast
– O(nlog2n), quick-sort
– Exponential: O(2n); very slow
Contributions from individual terms of f(n) = n2 + 100n
+ log10n + 1000 and how they varies with n
Formal definition of Big-O notation
• Definition: f(n) is O(g(n)) if there exist two
positive numbers c and N such the f(n) 
cg(n) for all n  N. The definition reads f(n)
is big-O of g, where g is an upper bound
on the value of f(n), meaning f grows at
most as fast as g as n increases.
Big-O notation: an example
• Let f(n) = 2n2 + 3n + 1
• Based on the definition, we can say f(n) = O(n2),
where g(n) = n2, if and only if we can find solutions for c and n such
that processor. Since c is positive, we can re-write the equation: 2 + 3 /
n + 1 / n2  c. There can be infinite set of solutions for the inequality.
For example, the above inequality is satisfied if c  6 and n = 1,
or c  3 ¾ and n = 2, etc., i.e., c = 6 and n = 1 pair is a solution; so is c
= 3 and n = 2.
• The best pair of (c, n) solution, we need to determine the smallest
value of n for which the dominant term (i.e., 2n2 in f(n)) becomes
larger than all other terms. Setting 2n2  3n, and 2n2 > 1, it is clear that
when n = 2, the two inequalities hold. From 2 + 3 / n + 1 / n2  c, we
obtain c  3 ¾.
Solutions to the inequality equation
2n2 + 3n + 1  cn2
Comparison of functions for different
values of c and n
Classes of algorithms and their execution
time (assuming processor speed =1 MIPS)
Typical function g(n) applied in big-O
estimates
Properties of big-O notation
• Transitivity: if f(n) is O(g(n)), and g(n) is O(h(n)),
then f(n) is O(h(n)).
• If f(n) is O(h(n)) and g(n) is O(h(n)) then f(n) +
g(n) is O(h(n)).
• The function ank is O(nk).
• The function nk is O(nk + j) for any positive j.
• If f(n) = cg(n), then f(n) is O(g(n)).
• The function logan is O(logbn) for any positive
numbers a and b  0.
The proofs are rather straightforward and appear on
p57 of textbook.
The  notation
• Definition: the function f(n) is  (f(n)) is there
exist positive numbers c and N such that f(n) 
cg(n) for all n  N. It reads "f is big-Omega of g"
and implies cg(n) is the lower bound of f(n).
• The above definition implies: f(n) is  (g(n)) if
and only if g(n) is O(f(n)).
• Similar to big-O situation, there can be infinite
number solutions for c and N pair. For practical
purposes, only the closest s are the interest,
which represents the largest lower bound.
The  (theta) notation
• Definition: f(n) is  (g(n)) if there exist
positive numbers c1, c2, and N such that
c1g(n)  f(n)  c2g(n) for all n  N. It reads
"f has an order of magnitude of g, or f is on
the order of g, or both f and g functions
grow at the same rate in the long run (i.e.,
when n continues to increase.)
Some observations
• Comparing two algorithms: O(n) and O(n2),
which one is faster?
• But, what about f(n) = 10 n2 + 108 n. Is it
O(n) or O(n2)? Obviously, for smaller value
of n (n  106) the second is faster. So,
sometimes it is desirable to consider
constants which are very large.
Additional examples
• for (i = sum = 0; i < n; i++)
sum += a[i];
// There are (2 + 2n) assignments (key ops); O(n)
• for (i = 0; i < n; i++){
for (j = 1, sum = a[0]; j <= i; j++)
sum += a[j];
cout << "Sum for sub-array 0 thru "
<< i << " is " << sum << endl;
}
// There are 1 + 3n + 2(1 + 2 + … + n – 1) = 1 + 3n + n(n –1)
// = O(n) + O(n2) = O(n2) assignment operations before the program is
completed.
Rigorous analyses of best,
average, and worst cases
• Best case: smallest number of steps.
• Worst case: maximum number of steps.
• Average case: the analysis may sometimes
be quite involved. There are a couple of
examples on pages 65 and 66.
The amortized complexity
• Takes into consideration of interdependence
between operations and their results. Will be
discussed in more detail in conjunction with
data structures.
as·ymp·tote (ăsʹĭm-tōt´, -ĭmp-) noun
Mathematics.
A line considered a limit to a curve in the
sense that the perpendicular distance from a
moving point on the curve to the line
approaches zero as the point moves an
infinite distance from the origin.
am·or·tize (ămʹər-tīz´, ə-môrʹ-) verb, transitive
am·or·tized, am·or·tiz·ing, am·or·tiz·es
1.
To liquidate (a debt, such as a
mortgage) by installment payments or
payment into a sinking fund.
2.
To write off an expenditure for (office
equipment, for example) by prorating over a
certain period.
Computational complexity
• Complexity of an algorithm measured in terms of
space and time.
– Space: main memory (RAM) needed to run the
program (implementation of the algorithm); less
important today because of the advancement of the
VLSI technology.
– Time: total execution time of an algorithms is not
meaningful because it is dependent on the hardware
(architecture or instruction set), on the language in
which the algorithm is code, and the efficiency of the
compiler (optimizer). A more meaningful metric is to
measure the number of operations as a function of size
n, which can be the number of elements in an array or a
data file.
The asymptotic behavior of math
functions
as·ymp·tote (ăsʹĭm-tōt´, -ĭmp-) noun
Mathematics.
A line considered a limit to a curve in the sense that
the perpendicular distance from a moving point on
the curve to the line approaches zero as the point
moves an infinite distance from the origin.
Examples: tan(), e-x, etc.
The asymptotic complexity: an example
• A simple loop that sum n elements in array a:
for (int i = sum = 0; i < n; n++)
sum += a[i];