In RobinHankin/disordR: Non-Ordered Vectors

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)
library("disordR")

Overview

Ordinary R vectors are unsuitable for working with values of associative maps because elements of an R vector may be accessed by reference to their location in the vector, but associative maps are stored in arbitrary order. However, when associating keys with values one needs both parts to be in 1-1 correspondence, so one cannot dispense with the order entirely. The disordR package includes a single S4 class, disord. This class allows one to perform only those operations appropriate for manipulating values (or keys) of associative maps.

A useful heuristic is that one is only allowed to access or modify a disord object using a python list comprehension. The idea is to prevent "illegal" operations on values (or keys) of associative maps, whose order is undefined or at best implementation-specific, while allowing sensible operations.

The `disord` package in use

I will illustrate the package by a few examples of legal and illegal code. First create a simple disord object:

set.seed(0)
a <- rdis()    # a random disord object
a

We may perform various operations on this object:

a+4
a > 5
a[a<6]

with no difficulty. But if we try to find elements of a with a particular offset or offsets, the system returns an error:

a[1]
a[c(2,3)]

This is because the elements of a are stored in an implementation-specific order and there is no such thing as the "first" element. We may manipulate elements of a by reference to their values but not by their position in the vector:

a[a<3] <- 0  # round small elements down
a

Replacement methods can access subsets where this makes sense:

x <- disord(1:10)
x
x[x<3] <- x[x<3] + 100
x

Two distinct `disord` objects

If we create another disord object, b:

b <- rdis()
b

Then a and b have the same length, but adding them vectorially is forbidden because the order of their elements is implementation-specific:

a+b

Also, replacement methods that access cross-referenced locations are forbidden:

a[b < 4] <- 5

(observe carefully that a[b<14] <- 5 is legal). However, sometimes one knows that two disord objects have the same ordering (perhaps a is the key, and b the value, of an associative array). In this case one may cross-reference them provided that the hash codes of the two objects agree:

a <- rdis()
b <- disord(sample(9),hash(a))
a
b

See how a and b have the same hash code. Then, although the elements of a and b are stored in an implementation-specific (and thus unknown) order, whatever order it is is the same in both objects and they are relatable:

a+b
a[b < 0.5]
a[b < 0.2] <- b[b < 0.2]
a

Comparison with python

A disord object is comparable with python arrays with the proviso that one is only allowed to access or manipulate elements via list comprehensions. See the following python session:

~ % python3
Python 3.9.5 (default, May  4 2021, 03:33:11) 
[Clang 12.0.0 (clang-1200.0.32.29)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from numpy import *
>>> a = random.rand(5)
>>> b = random.rand(5)
>>> a
array([0.13275574, 0.07243463, 0.0543058 , 0.36633955, 0.73795801])
>>> b
array([0.14934221, 0.11321413, 0.6135964 , 0.53558058, 0.6396702 ])
>>> [i**2 for i in a]
[0.017624086607707354, 0.005246776214236293, 0.002949120389839953, 0.1342046637421116, 0.54458202262916]
>>> [i+j for i,j in zip(a,b)]
[0.28209795105056, 0.18564876373593042, 0.6679022004516395, 0.901920128499422, 1.377628206483919]
>>>