Editing Selection algorithm (section)

==Algorithms==
===Sorting and heapselect===
As a baseline algorithm, selection of the {{nowrap|<math>k</math>th}} smallest value in a collection of values can be performed by the following two steps:
* [[Sorting|Sort]] the collection
* If the output of the sorting algorithm is an [[Array (data type)|array]], retrieve its {{nowrap|<math>k</math>th}} element; otherwise, scan the sorted sequence to find the {{nowrap|<math>k</math>th}} element.
The time for this method is dominated by the sorting step, which requires <math>\Theta(n\log n)</math> time using a {{nowrap|[[comparison sort]].{{r|clrs|skiena}}}} Even when [[integer sorting]] algorithms may be used, these are generally slower than the linear time that may be achieved using specialized selection algorithms. Nevertheless, the simplicity of this approach makes it attractive, especially when a highly-optimized sorting routine is provided as part of a runtime library, but a selection algorithm is not. For inputs of moderate size, sorting can be faster than non-random selection algorithms, because of the smaller constant factors in its running {{nowrap|time.{{r|erickson}}}} This method also produces a sorted version of the collection, which may be useful for other later computations, and in particular for selection with other choices {{nowrap|of <math>k</math>.{{r|skiena}}}}

For a sorting algorithm that generates one item at a time, such as [[selection sort]], the scan can be done in tandem with the sort, and the sort can be terminated once the {{nowrap|<math>k</math>th}} element has been found. One possible design of a consolation bracket in a [[single-elimination tournament]], in which the teams who lost to the eventual winner play another mini-tournament to determine second place, can be seen as an instance of this {{nowrap|method.{{r|bfprt}}}} Applying this optimization to [[heapsort]] produces the [[heapselect]] algorithm, which can select the {{nowrap|<math>k</math>th}} smallest value in {{nowrap|time <math>O(n+k\log n)</math>.{{r|brodal}}}} This is fast when <math>k</math> is small relative {{nowrap|to <math>n</math>,}} but degenerates to <math>O(n\log n)</math> for larger values {{nowrap|of <math>k</math>,}} such as the choice <math>k=n/2</math> used for median finding.

===Pivoting===
Many methods for selection are based on choosing a special "pivot" element from the input, and using comparisons with this element to divide the remaining <math>n-1</math> input values into two subsets: the set <math>L</math> of elements less than the pivot, and the set <math>R</math> of elements greater than the pivot. The algorithm can then determine where the {{nowrap|<math>k</math>th}} smallest value is to be found, based on a comparison of <math>k</math> with the sizes of these sets. In particular, {{nowrap|if <math>k\le|L|</math>,}} the {{nowrap|<math>k</math>th}} smallest value is {{nowrap|in <math>L</math>,}} and can be found recursively by applying the same selection algorithm {{nowrap|to <math>L</math>.}} {{nowrap|If <math>k=|L|+1</math>,}} then the {{nowrap|<math>k</math>th}} smallest value is the pivot, and it can be returned immediately. In the remaining case, the {{nowrap|<math>k</math>th}} smallest value is {{nowrap|in <math>R</math>,}} and more specifically it is the element in position <math>k-|L|-1</math> {{nowrap|of <math>R</math>.}} It can be found by applying a selection algorithm recursively, seeking the value in this position {{nowrap|in <math>R</math>.{{r|kletar}}}}

As with the related pivoting-based [[quicksort]] algorithm, the partition of the input into <math>L</math> and <math>R</math> may be done by making new collections for these sets, or by a method that partitions a given list or array data type in-place. Details vary depending on how the input collection is {{nowrap|represented.<ref>For instance, Cormen et al. use an in-place array partition, while Kleinberg and Tardos describe the input as a set and use a method that partitions it into two new sets.</ref>}} The time to compare the pivot against all the other values {{nowrap|is <math>O(n)</math>.{{r|kletar}}}} However, pivoting methods differ in how they choose the pivot, which affects how big the subproblems in each recursive call will be. The efficiency of these methods depends greatly on the choice of the pivot. If the pivot is chosen badly, the running time of this method can be as slow {{nowrap|as <math>O(n^2)</math>.{{r|erickson}}}}
*If the pivot were exactly at the median of the input, then each recursive call would have at most half as many values as the previous call, and the total times would add in a [[geometric series]] {{nowrap|to <math>O(n)</math>.}} However, finding the median is itself a selection problem, on the entire original input. Trying to find it by a recursive call to a selection algorithm would lead to an infinite recursion, because the problem size would not decrease in each {{nowrap|call.{{r|kletar}}}}
*[[Quickselect]] chooses the pivot uniformly at random from the input values. It can be described as a [[prune and search]] algorithm,{{r|gootam}} a variant of [[quicksort]], with the same pivoting strategy, but where quicksort makes two recursive calls to sort the two subcollections <math>L</math> {{nowrap|and <math>R</math>,}} quickselect only makes one of these two calls. Its [[expected time]] {{nowrap|is <math>O(n)</math>.{{r|clrs|kletar|gootam}}}} For any constant <math>C</math>, the probability that its number of comparisons exceeds <math>Cn</math> is superexponentially small {{nowrap|in <math>C</math>.{{r|devroye}}}}
*The [[Floyd–Rivest algorithm]], a variation of quickselect, chooses a pivot by randomly sampling a subset of <math>r</math> data values, for some sample {{nowrap|size <math>r</math>,}} and then recursively selecting two elements somewhat above and below position <math>rk/n</math> of the sample to use as pivots. With this choice, it is likely that <math>k</math> is sandwiched between the two pivots, so that after pivoting only a small number of data values between the pivots are left for a recursive call. This method can achieve an expected number of comparisons that is {{nowrap|<math>n+\min(k,n-k)+o(n)</math>.{{r|floriv}}}} In their original work, Floyd and Rivest claimed that the <math>o(n)</math> term could be made as small as <math>O(\sqrt n)</math> by a recursive sampling scheme, but the correctness of their analysis has been {{nowrap|questioned.{{r|brown|prt}}}} Instead, more rigorous analysis has shown that a version of their algorithm achieves <math>O(\sqrt{n\log n})</math> for this {{nowrap|term.{{r|knuth}}}} Although the usual analysis of both quickselect and the Floyd–Rivest algorithm assumes the use of a [[true random number generator]], a version of the Floyd–Rivest algorithm using a [[pseudorandom number generator]] seeded with only logarithmically many true random bits has been proven to run in linear time with high probability.{{r|karrag}}
[[File:Mid-of-mid.png|thumb|upright=1.35|Visualization of pivot selection for the [[median of medians]] method. Each set of five elements is shown as a column of dots in the figure, sorted in increasing order from top to bottom. If their medians (the green and purple dots in the middle row) are sorted in increasing order from left to right, and the median of medians is chosen as the pivot, then the <math>3n/10</math> elements in the upper left quadrant will be less than the pivot, and the <math>3n/10</math> elements in the lower right quadrant will be greater than the pivot, showing that many elements will be eliminated by pivoting.]]
*The [[median of medians]] method partitions the input into sets of five elements, and uses some other non-recursive method to find the median of each of these sets in constant time per set. It then recursively calls itself to find the median of these <math>n/5</math> medians. Using the resulting median of medians as the pivot produces a partition with {{nowrap|<math>\max(|L|,|R|)\le 7n/10</math>.}} Thus, a problem on <math>n</math> elements is reduced to two recursive problems on <math>n/5</math> elements (to find the pivot) and at most <math>7n/10</math> elements (after the pivot is used). The total size of these two recursive subproblems is at {{nowrap|most <math>9n/10</math>,}} allowing the total time to be analyzed as a geometric series adding {{nowrap|to <math>O(n)</math>.}} Unlike quickselect, this algorithm is deterministic, not {{nowrap|randomized.{{r|clrs|erickson|bfprt}}}} It was the first linear-time deterministic selection algorithm {{nowrap|known,{{r|bfprt}}}} and is commonly taught in undergraduate algorithms classes as an example of a [[Divide-and-conquer algorithm|divide and conquer]] that does not divide into two equal {{nowrap|subproblems.{{r|clrs|erickson|gootam|gurwitz}}}} However, the high constant factors in its <math>O(n)</math> time bound make it slower than quickselect in {{nowrap|practice,{{r|skiena|gootam}}}} and slower even than sorting for inputs of moderate {{nowrap|size.{{r|erickson}}}}
*Hybrid algorithms such as [[introselect]] can be used to achieve the practical performance of quickselect with a fallback to medians of medians guaranteeing worst-case <math>O(n)</math> {{nowrap|time.{{r|musser}}}}

===Factories===
The deterministic selection algorithms with the smallest known numbers of comparisons, for values of <math>k</math> that are far from <math>1</math> {{nowrap|or <math>n</math>,}} are based on the concept of ''factories'', introduced in 1976 by [[Arnold Schönhage]], [[Mike Paterson]], and {{nowrap|[[Nick Pippenger]].{{r|spp}}}} These are methods that build [[partial order]]s of certain specified types, on small subsets of input values, by using comparisons to combine smaller partial orders. As a very simple example, one type of factory can take as input a sequence of single-element partial orders, compare pairs of elements from these orders, and produce as output a sequence of two-element totally ordered sets. The elements used as the inputs to this factory could either be input values that have not been compared with anything yet, or "waste" values produced by other factories. The goal of a factory-based algorithm is to combine together different factories, with the outputs of some factories going to the inputs of others, in order to eventually obtain a partial order in which one element (the {{nowrap|<math>k</math>th}} smallest) is larger than some <math>k-1</math> other elements and smaller than another <math>n-k</math> others. A careful design of these factories leads to an algorithm that, when applied to median-finding, uses at most <math>2.942n</math> comparisons. For other values {{nowrap|of <math>k</math>,}} the number of comparisons is {{nowrap|smaller.{{r|dz99}}}}

===Parallel algorithms===
[[Parallel algorithm]]s for selection have been studied since 1975, when [[Leslie Valiant]] introduced the parallel comparison tree model for analyzing these algorithms, and proved that in this model selection using a linear number of comparisons requires <math>\Omega(\log\log n)</math> parallel steps, even for selecting the minimum or {{nowrap|maximum.{{r|valiant}}}} Researchers later found parallel algorithms for selection in <math>O(\log\log n)</math> steps, matching this {{nowrap|bound.{{r|akss|azapip}}}} In a randomized parallel comparison tree model it is possible to perform selection in a bounded number of steps and a linear number of {{nowrap|comparisons.{{r|reischuk}}}} On the more realistic [[parallel RAM]] model of computing, with exclusive read exclusive write memory access, selection can be performed in time <math>O(\log n)</math> with <math>O(n/\log n)</math> processors, which is optimal both in time and in the number of {{nowrap|processors.{{r|han}}}} With concurrent memory access, slightly faster parallel time is possible in {{nowrap|general,{{r|chr}}}} and the <math>\log n</math> term in the time bound can be replaced {{nowrap|by <math>\log k</math>.{{r|dieram}}}}

===Sublinear data structures===
When data is already organized into a [[data structure]], it may be possible to perform selection in an amount of time that is sublinear in the number of values. As a simple case of this, for data already sorted into an array, selecting the {{nowrap|<math>k</math>th}} element may be performed by a single array lookup, in constant {{nowrap|time.{{r|frejoh}}}} For values organized into a two-dimensional array of {{nowrap|size <math>m\times n</math>,}} with sorted rows and columns, selection may be performed in time {{nowrap|<math>O\bigl(m\log(2n/m)\bigr)</math>,}} or faster when <math>k</math> is small relative to the array {{nowrap|dimensions.{{r|frejoh|kkzz}}}} For a collection of <math>m</math> one-dimensional sorted arrays, with <math>k_i</math> items less than the selected item in the {{nowrap|<math>i</math>th}} array, the time is {{nowrap|<math display=inline>O\bigl(m+\sum_{i=1}^m\log(k_i+1)\bigr)</math>.{{r|kkzz}}}}

Selection from data in a [[binary heap]] takes {{nowrap|time <math>O(k)</math>.}} This is independent of the size <math>n</math> of the heap, and faster than the <math>O(k\log n)</math> time bound that would be obtained from {{nowrap|[[best-first search]].{{r|kkzz|frederickson}}}} This same method can be applied more generally to data organized as any kind of heap-ordered tree (a tree in which each node stores one value in which the parent of each non-root node has a smaller value than its child). This method of performing selection in a heap has been applied to problems of listing multiple solutions to combinatorial optimization problems, such as finding the [[k shortest path routing|{{mvar|k}} shortest paths]] in a weighted graph, by defining a [[State space (computer science)|state space]] of solutions in the form of an [[implicit graph|implicitly defined]] heap-ordered tree, and then applying this selection algorithm to this {{nowrap|tree.{{r|kpaths}}}} In the other direction, linear time selection algorithms have been used as a subroutine in a [[priority queue]] data structure related to the heap, improving the time for extracting its {{nowrap|<math>k</math>th}} item from <math>O(\log n)</math> to {{nowrap|<math>O(\log^* n+\log k)</math>;}} here <math>\log^* n</math> is the {{nowrap|[[iterated logarithm]].{{r|bks}}}}

For a collection of data values undergoing dynamic insertions and deletions, the [[order statistic tree]] augments a [[self-balancing binary search tree]] structure with a constant amount of additional information per tree node, allowing insertions, deletions, and selection queries that ask for the {{nowrap|<math>k</math>th}} element in the current set to all be performed in <math>O(\log n)</math> time per {{nowrap|operation.{{r|clrs}}}} Going beyond the comparison model of computation, faster times per operation are possible for values that are small integers, on which binary arithmetic operations are {{nowrap|allowed.{{r|pattho}}}} It is not possible for a [[streaming algorithms|streaming algorithm]] with memory sublinear in both <math>n</math> and <math>k</math> to solve selection queries exactly for dynamic data, but the [[count–min sketch]] can be used to solve selection queries approximately, by finding a value whose position in the ordering of the elements (if it were added to them) would be within <math>\varepsilon n</math> steps of <math>k</math>, for a sketch whose size is within logarithmic factors of <math>1/\varepsilon</math>.{{r|cormut}}