---
title: "Performance"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Performance}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
```

Accuracy and flexibility are among the strong suits of moder. Performance is not. `mode_all()` and friends are multiple times slower than R's built-in functions for common measures of central tendency, `mean()` and `median()`.

Why is that? All of moder's code is written in R. By contrast, the default `mean()` method calls an internal function. Also, handling missing values is much more complex in the mode functions than in `mean()` or `median()`; see `vignette("missings")`. The default methods for `mean()` and `median()` are fairly short when compared to `mode_first()`, and to `mode_all()` if its helper function is included. `mode_single()` is a wrapper around `mode_all()`, so it's even longer.

It is also possible that some of the code is less efficient than it could be. In particular, rewriting moder's R code in a language like C++ or Rust might blend accuracy with high performance. I myself am not going to port the package in the near term, so help is welcome. If you have any suggestions, please [open an issue](https://github.com/lhdjung/moder/issues) or get in touch via email (jung-lukas\@gmx.net).

```{r, eval=FALSE, include=FALSE}
library(moder)
library(profvis)
library(bench)
library(ggplot2)

x <- c(2, 4, 6, NA, 3, 1, 1, 8, 1, 7)

# Profiling `mode_first()`:
na.rm <- FALSE
accept <- FALSE

profvis(purrr::map(
  .x = rep(list(x), 100000),
  .f = mode_first
))

# Benchmarking:
df <- mark(
  mode_single = mode_single(x),
  mode_first = mode_first(x),
  mode_all = mode_all(x),
  mean = mean(x),
  median = median(x),
  check = FALSE, iterations = 100000
)
```