Competitive analysis (online algorithm)
Encyclopedia
Competitive analysis is a method invented for analyzing online algorithm
s, in which the performance of an online algorithm (which must satisfy an unpredictable sequence of requests, completing each request without being able to see the future) is compared to the performance of an optimal offline algorithm that can view the sequence of requests in advance. An algorithm is competitive if its competitive ratio—the ratio between its performance and the offline algorithm's performance—is bounded. Unlike traditional worst-case analysis
, where the performance of an algorithm is measured only for "hard" inputs, competitive analysis requires that an algorithm perform well both on hard and easy inputs, where "hard" and "easy" are defined by the performance of the optimal offline algorithm.
For many algorithms, performance is dependent not only on the size of the inputs, but also on their values. One such example is the quicksort algorithm, which sorts an array of elements. Such data-dependent algorithms are analysed for average-case and worst-case data. Competitive analysis is a way of doing worst case analysis for on-line and randomized algorithm
s, which are typically data dependent.
In competitive analysis, one imagines an "adversary" that deliberately chooses difficult data, to maximize the ratio of the cost of the algorithm being studied and some optimal algorithm. Adversaries range in power from the oblivious adversary, which has no knowledge of the random choices made by the algorithm pitted against it, to the adaptive adversary that has full knowledge of how an algorithm works and its internal state at any point during its execution. Note that this distinction is only meaningful for randomized algorithms. For a deterministic algorithm, either adversary can simply compute what state that algorithm must have at any time in the future, and choose difficult data accordingly.
For example, the quicksort algorithm chooses one element, called the "pivot", that is, on average, not too far from the center value of the data being sorted. Quicksort then separates the data into two piles, one of which contains all elements with value less than the value of the pivot, and the other containing the rest of the elements. If quicksort chooses the pivot in some deterministic fashion (for instance, always choosing the first element in the list), then it is easy for an adversary to arrange the data beforehand so that quicksort will perform in worst-case time. If, however, quicksort chooses some random element to be the pivot, then an adversary without knowledge of what random numbers are coming up cannot arrange the data to guarantee worst-case execution time for quicksort.
The classic on-line problem first analysed with competitive analysis is the List update problem: Given a list of items and a sequence of requests for the various items, minimize the cost of accessing the list where the elements closer to the front of the list cost less to access. (Typically, the cost of accessing an item is equal to its position in the list.) After an access, the list may be rearranged. Most rearrangements have a cost. The Move-To-Front algorithm simply moves the requested item to the front after the access, at no cost. The Transpose algorithm swaps the accessed item with the item immediately before it, also at no cost. Classical methods of analysis showed that Transpose is optimal in certain contexts. In practice, Move-To-Front performed much better. Competitive analysis was used to show that an adversary can make Transpose perform arbitrarily badly compared to an optimal algorithm, whereas Move-To-Front can never be made to incur more than twice the cost of an optimal algorithm.
In the case of online requests from a server, competitive algorithms are used to overcome uncertainties about the future. That is, the algorithm does not "know" the future, while the imaginary adversary (the "competitor") "knows". Similarly, competitive algorithms were developed for distributed systems, where the algorithm has to react to a request arriving at one location, without "knowing" what has just happened in a remote location. This setting was presented in .
Online algorithm
In computer science, an online algorithm is one that can process its input piece-by-piece in a serial fashion, i.e., in the order that the input is fed to the algorithm, without having the entire input available from the start. In contrast, an offline algorithm is given the whole problem data from...
s, in which the performance of an online algorithm (which must satisfy an unpredictable sequence of requests, completing each request without being able to see the future) is compared to the performance of an optimal offline algorithm that can view the sequence of requests in advance. An algorithm is competitive if its competitive ratio—the ratio between its performance and the offline algorithm's performance—is bounded. Unlike traditional worst-case analysis
Best, worst and average case
In computer science, best, worst and average cases of a given algorithm express what the resource usage is at least, at most and on average, respectively...
, where the performance of an algorithm is measured only for "hard" inputs, competitive analysis requires that an algorithm perform well both on hard and easy inputs, where "hard" and "easy" are defined by the performance of the optimal offline algorithm.
For many algorithms, performance is dependent not only on the size of the inputs, but also on their values. One such example is the quicksort algorithm, which sorts an array of elements. Such data-dependent algorithms are analysed for average-case and worst-case data. Competitive analysis is a way of doing worst case analysis for on-line and randomized algorithm
Randomized algorithm
A randomized algorithm is an algorithm which employs a degree of randomness as part of its logic. The algorithm typically uses uniformly random bits as an auxiliary input to guide its behavior, in the hope of achieving good performance in the "average case" over all possible choices of random bits...
s, which are typically data dependent.
In competitive analysis, one imagines an "adversary" that deliberately chooses difficult data, to maximize the ratio of the cost of the algorithm being studied and some optimal algorithm. Adversaries range in power from the oblivious adversary, which has no knowledge of the random choices made by the algorithm pitted against it, to the adaptive adversary that has full knowledge of how an algorithm works and its internal state at any point during its execution. Note that this distinction is only meaningful for randomized algorithms. For a deterministic algorithm, either adversary can simply compute what state that algorithm must have at any time in the future, and choose difficult data accordingly.
For example, the quicksort algorithm chooses one element, called the "pivot", that is, on average, not too far from the center value of the data being sorted. Quicksort then separates the data into two piles, one of which contains all elements with value less than the value of the pivot, and the other containing the rest of the elements. If quicksort chooses the pivot in some deterministic fashion (for instance, always choosing the first element in the list), then it is easy for an adversary to arrange the data beforehand so that quicksort will perform in worst-case time. If, however, quicksort chooses some random element to be the pivot, then an adversary without knowledge of what random numbers are coming up cannot arrange the data to guarantee worst-case execution time for quicksort.
The classic on-line problem first analysed with competitive analysis is the List update problem: Given a list of items and a sequence of requests for the various items, minimize the cost of accessing the list where the elements closer to the front of the list cost less to access. (Typically, the cost of accessing an item is equal to its position in the list.) After an access, the list may be rearranged. Most rearrangements have a cost. The Move-To-Front algorithm simply moves the requested item to the front after the access, at no cost. The Transpose algorithm swaps the accessed item with the item immediately before it, also at no cost. Classical methods of analysis showed that Transpose is optimal in certain contexts. In practice, Move-To-Front performed much better. Competitive analysis was used to show that an adversary can make Transpose perform arbitrarily badly compared to an optimal algorithm, whereas Move-To-Front can never be made to incur more than twice the cost of an optimal algorithm.
In the case of online requests from a server, competitive algorithms are used to overcome uncertainties about the future. That is, the algorithm does not "know" the future, while the imaginary adversary (the "competitor") "knows". Similarly, competitive algorithms were developed for distributed systems, where the algorithm has to react to a request arriving at one location, without "knowing" what has just happened in a remote location. This setting was presented in .
See also
- Adversary (online algorithm)Adversary (online algorithm)In computer science, an online algorithm measures its competitiveness against different adversary models. For deterministic algorithms, the adversary is the same, the adaptive offline adversary...
- Amortized analysisAmortized analysisIn computer science, amortized analysis is a method of analyzing algorithms that considers the entire sequence of operations of the program. It allows for the establishment of a worst-case bound for the performance of an algorithm irrespective of the inputs by looking at all of the operations...
- K-server problemK-server problemThe k-server problem is a problem of theoretical computer science in the category of online algorithms, one of two abstract problems on metric spaces that are central to the theory of competitive analysis...
- Online algorithmOnline algorithmIn computer science, an online algorithm is one that can process its input piece-by-piece in a serial fashion, i.e., in the order that the input is fed to the algorithm, without having the entire input available from the start. In contrast, an offline algorithm is given the whole problem data from...
- List update problem