In jr-packages/jrAdvPackage: Course Material for Advanced R / Building an R Package Combo

Advanced R programming: Practical 2

Argument matching

R allows a variety of ways to match function arguments. For example, by position, by complete name, or by partial name. We didn't cover argument matching in the lecture, so let's try and figure out the rules from the examples below. First we'll create a little function to help

arg_explore = function(arg1, rg2, rg3)
    paste("a1, a2, a3 = ", arg1, rg2, rg3)

\noindent Next we'll create a few examples. Try and predict what's going to happen before calling the functions. One of these examples will raise an error - why?

arg_explore(1, 2, 3)
arg_explore(2, 3, arg1 = 1)
arg_explore(2, 3, a = 1)
arg_explore(1, 3, rg = 1)

\noindent Can you write down a set of rules that R uses when matching arguments?

## SOLUTION
## See http://goo.gl/NKsved for the offical document
## To summeriase, matching happens in a three stage pass:
#1. Exact matching on tags
#2. Partial matching on tags.
#3. Positional matching

Following on from the above example, can you predict what will happen with

plot(type = "l", 1:10, 11:20)

\noindent and

rnorm(mean = 4, 4, n = 5)

## SOLUTION
#plot(type="l", 1:10, 11:20) is equivilent to
plot(x = 1:10, y = 11:20, type = "l")
#rnorm(mean=4, 4, n=5) is equivilent to
rnorm(n = 5, mean = 4, sd = 4)

Functions as first class objects

Suppose we have a function that performs a statistical analysis

## Use regression as an example
stat_ana = function(x, y) {
  lm(y ~ x)
}

\noindent However, we want to alter the input data set using different transformations. For example, the $\log$ transformation. In particular, we want the ability to pass arbitrary transformation functions to stat_ana.

• Add an argument trans to the stat_ana() function. This argument should have a default value of NULL.

## SOLUTION
stat_ana = function(x, y, trans=NULL) {
  lm(y ~ x)
}

• Use is.function() to test whether a function has been passed to trans, transform the vectors x and y when appropriate. For example,

stat_ana(x, y, trans = log)

\noindent would take log's of x and y.

## SOLUTION
stat_ana = function(x, y, trans=NULL) {
  if (is.function(trans)) {
    x = trans(x)
    y = trans(y)
  }
  lm(y ~ x)
}

• Allow the trans argument to take character arguments in additional to function arguments. For example, if we used trans = 'normalise', then we would normalise the data i.e. subtract the mean and divide by the standard deviation.

## SOLUTION
stat_ana = function(x, y, trans=NULL) {
  if (is.function(trans)) {
    x = trans(x)
    y = trans(y)
  } else if (trans == "normalise") {
    x = scale(x)
    y = scale(y)
  }
  lm(y ~ x)
}

Variable scope

Scoping can get tricky. Before running the example code below, predict what is going to happen

1. A simple one to get started

f = function(x) return(x + 1)
f(10)

##Nothing strange here. We just get
f(10)

1. A bit more tricky

f = function(x) {
  f = function(x) {
    x + 1
  }
  x = x + 1
  return(f(x))
}
f(10)

1. More complex

f = function(x) {
  f = function(x) {
    f = function(x) {
      x + 1
    }
    x = x + 1
    return(f(x))
  }
  x = x + 1
  return(f(x))
}
f(10)

## Solution: The easiest way to understand is to use print statements
f = function(x) {
  f = function(x) {
    f = function(x) {
      message("f1: = ", x)
      x + 1
    }
    message("f2: = ", x)
    x = x + 1
    return(f(x))
  }
  message("f3: = ", x)
  x = x + 1
  return(f(x))
}
f(10)

f = function(x) {
  f = function(x) {
    x = 100
    f = function(x) {
      x + 1
    }
    x = x + 1
    return(f(x))
  }
  x = x + 1
  return(f(x))
}
f(10)

##Solution: The easiest way to understand is to use print statements as above

Function closures

Following the examples in the notes, where we created a function closure for the normal and uniform distributions. Create a similar closure for

• the Poisson distribution,\sidenote{Hint: see rpois and dpois.}

poisson = function(lambda) {
     r = function(n=1) rpois(n, lambda)
     d = function(x, log=FALSE) dpois(x, lambda, log = log)
     return(list(r = r, d = d))
}

• and the Geometric distribution.\sidenote{Hint: see rgeom and dgeom.}

geometric = function(prob) {
     r = function(n=1) rgeom(n, prob)
     d = function(x, log=FALSE) dgeom(x, prob, log = log)
     return(list(r = r, d = d))
}

Multiple column types

In the below code, I've attempted to loop through a data frame and extract the maximum values.

dd = data.frame(w = rnorm(10), 
                x = letters[1:10],
                y = rnorm(10),
                z = rnorm(10)
)

max_cols = rep(NA, ncol(dd))
for (i in seq_along(dd)) {
  max_cols[i] = max(dd[, i])
}
max_cols

\noindent However, there's something wrong. The second column isn't numeric and so the for loop breaks when we get to there. Of course, we could just change the iterations to c(1,3,4) to leave out the second column. But imagine we have tens of columns. Use the try() function to bypass the error.

dd = data.frame(w = rnorm(10), 
                x = letters[1:10],
                y = rnorm(10),
                z = rnorm(10)
)

max_cols = rep(NA, ncol(dd))
for (i in seq_along(dd)) {
  try(max_cols[i] <- max(dd[, i]))
}
max_cols

Solutions

Solutions are contained within the course package

library("jrAdvPackage")
vignette("solutions2", package = "jrAdvPackage")

jr-packages/jrAdvPackage documentation built on Dec. 27, 2019, 8:05 a.m.