The next set of tutorials focuses on more advanced functionality in distr6. This is not to say that the code itself may be any more complex, but that the use-cases for this functionality is more advanced. Examples of these include composite distributions, decorators for numeric results and piping methods. We start with S3 and piping as these (especially S3) will be familiar concepts to most users, are available in distr6 but are not part of the core functionality.
Using the R62S3 package (Sonabend 2019)
most R6 methods in distr6 can instead be called as an S3 method. Even if
you’re not aware of it, you’ve probably used S3 every time you’ve opened
R. For example the plot function produces a different plot
depending on what object it’s called on, a numeric, a linear regression
model, or something more complicated, this is because of the S3 dispatch
system. Every distribution, kernel, parameter set, wrapper and decorator
method is available through S3. Note that constructors always use R6
syntax, i.e. $new(). Below are some examples, see the
documentation help pages for the different S3 and R6 syntax but remember
these are almost identical except that the first argument in S3 will be
the object whereas this is omitted in the R6 usage.
You can select which classes to create methods for with R62S3, below
we are just interested in methods associated with the
Distribution class, but this will work with any R6 class in
{distr6}.
## generate methods from the public methods and active bindings of `Distribution`
R62S3::R62Fun(Distribution, assignEnvir = topenv(), scope = c("public", "active"))
N <- Normal$new()
N$pdf(1:2) # R6
#> [1] 0.24197072 0.05399097
pdf(N, 1:2) # S3
#> [1] 0.24197072 0.05399097
N$summary()
#> Normal Probability Distribution.
#> Parameterised with:
#>
#> Id Support Value Tags
#> <char> <char> <list> <list>
#> 1: mean ℝ 0 required
#> 2: prec ℝ+ linked,required
#> 3: sd ℝ+ linked,required
#> 4: var ℝ+ 1 linked,required
#>
#>
#> Quick Statistics
#> Mean: 0
#> Variance: 1
#> Skewness: 0
#> Ex. Kurtosis: 0
#>
#> Support: ℝ Scientific Type: ℝ
#>
#> Traits: continuous; univariate
#> Properties: symmetric; mesokurtic; no skew
summary(N)
#> Normal Probability Distribution.
#> Parameterised with:
#>
#> Id Support Value Tags
#> <char> <char> <list> <list>
#> 1: mean ℝ 0 required
#> 2: prec ℝ+ linked,required
#> 3: sd ℝ+ linked,required
#> 4: var ℝ+ 1 linked,required
#>
#>
#> Quick Statistics
#> Mean: 0
#> Variance: 1
#> Skewness: 0
#> Ex. Kurtosis: 0
#>
#> Support: ℝ Scientific Type: ℝ
#>
#> Traits: continuous; univariate
#> Properties: symmetric; mesokurtic; no skew
setParameterValue(N,var = 2)
getParameterValue(N, "var")
#> [1] 2We won’t go into too much detail about piping because it will be
familiar to users who need to make use of it and may not be necessary
for others. Piping is the process of calling functions in a set order
using the pipe, %>% notation, from the magrittr package
(Milton Bache and Wickham 2014). We
specifically mention piping because it goes hand-in-hand with method
chaining, which we discussed in the introduction
to R6 tutorial. Below are some examples of how piping can be used in
distr6.
library(magrittr)
Normal$new() %>% setParameterValue(var = 2) %>% getParameterValue("var")
#> [1] 2
## generate methods from the public methods and active bindings of `ExoticStatistics`
R62S3::R62Fun(ExoticStatistics, assignEnvir = topenv(), scope = c("public", "active"))
# Don't worry about unfamiliar functions used here, we return to this in the next tutorial
Normal$new() %>% decorate("ExoticStatistics") %>% hazard(2)
#> Normal is now decorated with ExoticStatistics
#> [1] 2.373216Notice that usually the S3 methods require the first argument to be
the object, in this case N or Normal$new() but
by using the pipe notation, this is filled automatically. For a more
in-depth tutorial on magrittr see here.
In this tutorial we covered using S3 and piping instead of/as well as R6 syntax in distr6. These allow you to use whichever syntax you’re most comfortable with whilst maintaining the integrity of a well-designed and (hopefully) user-friendly object-oriented interface. In the next tutorial we look at another big feature in distr6, decorators.