comboGroups {RcppAlgos}

Partition a Vector into Groups

Description

• Generate partitions of a vector into groups. See Create Combinations in R by Groups on https://stackoverflow.com for a direct use case of when the groups sizes are equal.

• Produce results in parallel using the Parallel or nThreads arguments.

• GMP support allows for exploration where the number of results is large.

• The output is in lexicographical order by groups.

Usage

comboGroups(v, numGroups = NULL, grpSizes = NULL,
            retType = "matrix", lower = NULL, upper = NULL,
            Parallel = FALSE, nThreads = NULL)

Arguments

v

Source vector. If v is a positive integer, it will be converted to the sequence 1:v. If v is a negative integer, it will be converted to the sequence v:-1. All atomic types are supported (See is.atomic).

numGroups

An Integer. The number of groups that the vector will be partitioned into. The default is NULL. If provided and grpSize is NULL, it must divide the length of v (if v is a vector) or v (if v is a scalar).

grpSizes

A vector of whole numbers representing the size of each group. The default is NULL. If provided, the sum of the elements must total the length of v (if v is a vector) or v (if v is a scalar).

retType

A string, "3Darray" or "matrix", that determines the shape of the output. The default is "matrix". Note, "3Darray" can only be used when the size of each group is uniform. When the size of each group varies, the return output will always be a matrix.

lower

The lower bound. Partitions of groups are generated lexicographically, thus utilizing this argument will determine which specific result to start generating from (e.g. comboGroups(8, 2, lower = 30) is equivalent to comboGroups(8, 2)[30:comboGroupsCount(8, 2), ]). This argument along with upper is very useful for generating results in chunks allowing for easy parallelization.

upper

The upper bound. Similar to lower, however this parameter allows the user to stop generation at a specific result (e.g. comboGroups(8, 2, upper = 5) is equivalent to comboGroups(8, 2)[1:5, ])

Parallel

Logical value indicating whether results should be generated in parallel using n - 1 threads, where n is the maximum number of threads. The default is FALSE. If nThreads is not NULL, it will be given preference (e.g. if user has 8 threads with Parallel = TRUE and nThreads = 4, only 4 threads will be spawned). If your system is single-threaded, the arguments Parallel and nThreads are ignored.

nThreads

Specific number of threads to be used. The default is NULL. See Parallel.

Details

Conceptually, this problem can be viewed as generating all permutations of the vector v and removing the within group permutations. To illustrate this, let us consider the case of generating partitions of 1:8 into 2 groups each of size 4.

• To begin, generate the permutations of 1:8 and group the first/last four elements of each row.

  			Grp1			Grp2
  	C1	C2	C3	C4	C5	C6	C7	C8
  R1	| 1	2	3	4 |	| 5	6	7	8 |
  R2	| 1	2	3	4 |	| 5	6	8	7 |
  R3	| 1	2	3	4 |	| 5	7	6	8 |
  R4	| 1	2	3	4 |	| 5	7	8	6 |
  R5	| 1	2	3	4 |	| 5	8	6	7 |
  R6	| 1	2	3	4 |	| 5	8	7	6 |

• Note that the permutations above are equivalent partitions of 2 groups of size 4 as only the last four elements are permuted. If we look at at the 25^{th} lexicographical permutation, we observe our second distinct partition.

  			Grp1			Grp2
  	C1	C2	C3	C4	C5	C6	C7	C8
  R24	| 1	2	3	4 |	| 8	7	6	5 |
  R25	| 1	2	3	5 |	| 4	6	7	8 |
  R26	| 1	2	3	5 |	| 4	6	8	7 |
  R27	| 1	2	3	5 |	| 4	7	6	8 |
  R28	| 1	2	3	5 |	| 4	7	8	6 |

• Continuing on, we will reach the 3,457^{th} lexicographical permutation, which represents the last result:

  			Grp1			Grp2
  	C1	C2	C3	C4	C5	C6	C7	C8
  R3454	| 1	6	7	5 |	|8	3	4	2 |
  R3455	| 1	6	7	5 |	|8	4	2	3 |
  R3456	| 1	6	7	5 |	|8	4	3	2 |
  R3457	| 1	6	7	8 |	| 2	3	4	5 |
  R3458	| 1	6	7	8 |	|2	3	5	4 |

• For this small example, the method above will not be that computationally expensive. In fact, there are only 35 total partitions of 1:8 into 2 groups of size 4 out of a possible factorial(8) = 40320 permutations. However, just doubling the size of the vector will make this approach infeasible as there are over 10 trillion permutations of 1:16.

• The algorithm in comboGroups avoids these duplicate partitions of groups by utilizing an efficient algorithm analogous to the std::next_permutation found in the standard algorithm library in C++.

Value

By default, a matrix is returned with column names corresponding to the associated group. If retType = "3Darray", a named 3D array is returned.

Note

• The maximum number of partitions of groups that can be generated at one time is 2^{31} - 1. Utilizing lower and upper makes it possible to generate additional combinations/permutations.

• The length of grpSizes must equal numGroups if both grpSize and numGroups are provided.

Author(s)

Joseph Wood

Examples

## return a matrix
comboGroups(8, 2)

## or a 3 dimensional array
temp = comboGroups(8, 2, retType = "3Darray")

## view the first partition
temp[1, , ]

## Example with groups of varying size
comboGroups(8, grpSizes = c(3, 5))

total = comboGroupsCount(11, grpSizes = c(3, 3, 5))

## Start generating from particular index
comboGroups(11, grpSizes = c(3, 3, 5), lower = total - 20)