JonasMoss

#' Evaluates a call as if its function was defined in a specified environment
#' 
#' When a name is encountered in the definition of a function, the search path
#' for that name is given by the defining environment of the function. This is
#' good behaviour, since it allows simple reasoning about how a function should
#' behave: If two calls to a function defined in a constant environment \code{e} 
#' yield different results, this must be because they are given different 
#' arguments. 
#' 
#' Sometimes, a function is defined to make messy code more readable, but is 

JonasMoss

#' Allow self-referential arguments in functions.
#' 
#' @param call A function call.
#' @param quote Logical; if \code{TRUE}, the supplied \code{call} is interpreted
#' as a quote, so \code{substitute} is applied.
#' @return The evaluated function call with the self-refering arguments 
#' evaluated.
#' @examples 
#'     R(plot(y = 1:10, x = y^2))
#'     R(plot(x = y^2, y = 1:10))

JonasMoss

## A small simulation of how powers could be distributed in psychology.
## 'sn' is the skew-normal distribution, which I suppose is useful in this case.
## The package is available from CRAN: 
## install.packages("sn")

set.seed(313)
N = 100000
thetas = sn::rsn(N, xi = 0.05, omega = 0.15, alpha = 2) # Sample of true thetas.

## I assume the effect sizes (thetas) are sampled from the following 

JonasMoss

## A small demonstration of statistical power for a one-sided z-test. The point 
## is to show that powers close to zero or one are not reasonable to assume a 
## priori. 

n = 80
theta = seq(-0.3, 1, by = 0.01)
plot(theta, 1 - pnorm(1.96 - sqrt(n)*theta), type = "l", bty = "l",
     xlab = expression(theta), ylab = "Power",
     main = paste0("Power when n = ", n))
grid()

JonasMoss

#' Create an \code{alist}.
#'
#' An alternative to \code{base::alist} that works with \code{...} objects
#' inside a function.
#'
#' An \code{alist} is a list containing possibly unevaluated arguments. The
#' standard implementation of \code{alist} does not work with \code{...} objects
#' inside a function, but this one does. Uses of this function includes partial
#' function application and easier handling of generics that make use of
#' \code{...}, for instance \code{plot}.

JonasMoss

n = 30

param_loo = vector("list", n)
print(length(param_loo)) 
# [1] 30

for (i in 1:n) {
  param_loo[[i]] = NULL
}

JonasMoss

## These are generic functions for extracting and assigning values to the enclosing
## environment of a function. It needs class(f) = "function" to work.

`$.function` = function(f, y) environment(f)[[y]]

`$<-.function` = function(f, y, value) {
  environment(f)[[y]] = value
  invisible(f)
}

JonasMoss

## A simple example of a functor in R.

## A simple function of one argument. Since y is neither defined inside the function body or
## given as an argument, R will search the enclosing environment for y. Since R is lazy, it 
## won't do this when the function is defined, but only when the function is called (and y
## is needed.)
f = function(x) x^2 + y^2

## This changes the enclosing environment of f to a new environment. It was the global
## environment before, and we don't want that. 

JonasMoss

## Functor example 1: Memoization. 
## We memoize a function g with up to N values.

## The function to memoize: It is slow to calculate, so we'll gain from 
## storing its values.

g = function(x) {
    Sys.sleep(0.1)
    print("Processesing ...")
    x^2

JonasMoss

## Make a function rkde that samples from a kernel density

#' Sample from a kernel density.
#' 
#' @param n Number of observations to sample.
#' @param x The data from which the estimate is to be computed.
#' @param bw Desired bandwidth.
#' @return A numeric vector with n sampled data points from the kernel
#' density estimator.