Skip to contents

pr

pr provides a set of probability classes built on vctrs and its rcrds class. The primary offerings include:

  • prob: A probability vector class
  • prob_mat: A rcrd which describes a matrix of probabilities
  • cnt_prob_mat: A rcrd which describes a matrix of probabilities with counts

It is designed to be lightweight so that it can be included as a dependency for other packages. Its only direct dependency is vctrs.

Installation

You can install the development version of pr from GitHub with:

# install.packages("pak")
pak::pak("christopherkenny/pr")

Example

The primary component of the package is the prob class. This represents a vector of probabilities.

library(pr)
prob(x = c(1 / 2, 0.4, NA, 0.875))
#> <probability[4]>
#> [1] 0.500 0.400 <NA>  0.875

It supports creating record classes for sets of probabilities which sum to 1 row-wise. This is implemented via the vctrs rcrd type, which is a collection of same-length vectors.

A corresponding probability matrix can be created as follows:


p1 <- prob(c(0.4, 2 / 3, NA_real_, 1L))
p2 <- 1 - p1
prob_mat(pr_one = p1, pr_two = p2)
#> <probability matrix[4]>
#>      pr_one pr_two
#> [1,] 0.400  0.600 
#> [2,] 0.667  0.333 
#> [3,] <NA>   <NA>  
#> [4,] 1.000  0.000

Finally, a count probability matrix is provided which adds a count field to the probability matrix.

p1 <- prob(c(0.4, 2 / 3, NA_real_, 1L))
p2 <- 1 - p1
cnt_prob_mat(pr_one = p1, pr_two = p2, count = c(10, 20, 5, 15))
#> <count probability matrix[4]>
#>      count pr_one pr_two
#> [1,]    10 0.400  0.600 
#> [2,]    20 0.667  0.333 
#> [3,]     5 <NA>   <NA>  
#> [4,]    15 1.000  0.000

These can also be created without specifying the count, in which case it defaults to 1L.

p1 <- prob(c(0.4, 2 / 3, NA_real_, 1L))
p2 <- 1 - p1
cnt_prob_mat(pr_one = p1, pr_two = p2)
#> <count probability matrix[4]>
#>      count pr_one pr_two
#> [1,]     1 0.400  0.600 
#> [2,]     1 0.667  0.333 
#> [3,]     1 <NA>   <NA>  
#> [4,]     1 1.000  0.000

For both matrix types, these are represented as rcrd classes which require the columns to be named and each column to be a prob vector (or coercible to one).