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

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

Using `tidylog` adds a small overhead to each function call. For instance, because tidylog needs
to figure out how many rows were dropped when you use `tidylog::filter`,
this call will be a bit slower than using `dplyr::filter` directly.
The overhead is usually not noticeable, but can be for larger datasets, especially when using joins.
The benchmarks below give some impression of how large the overhead is.


```{r message = FALSE, warning = FALSE}
library("dplyr")
library("tidylog", warn.conflicts = FALSE)
library("bench")
library("knitr")
```

## filter

On a small dataset:

```{r message = FALSE}
bench::mark(
    dplyr::filter(mtcars, cyl == 4),
    tidylog::filter(mtcars, cyl == 4), iterations = 100
) %>%
    dplyr::select(expression, min, median, n_itr) %>%
    kable()
```

On a larger dataset:

```{r message = FALSE}
df <- tibble(x = rnorm(100000))

bench::mark(
    dplyr::filter(df, x > 0),
    tidylog::filter(df, x > 0), iterations = 100
) %>%
    dplyr::select(expression, min, median, n_itr) %>%
    kable()
```

## mutate

On a small dataset:

```{r message = FALSE}
bench::mark(
    dplyr::mutate(mtcars, cyl = as.factor(cyl)),
    tidylog::mutate(mtcars, cyl = as.factor(cyl)), iterations = 100
) %>%
    dplyr::select(expression, min, median, n_itr) %>%
    kable()
```

On a larger dataset:

```{r message = FALSE}
df <- tibble(x = round(runif(10000) * 10))

bench::mark(
    dplyr::mutate(df, x = as.factor(x)),
    tidylog::mutate(df, x = as.factor(x)), iterations = 100
) %>%
    dplyr::select(expression, min, median, n_itr) %>%
    kable()
```

## joins

Joins are the most expensive operation, as tidylog has to do
two additional joins behind the scenes.

On a small dataset:

```{r message = FALSE}
bench::mark(
    dplyr::inner_join(band_members, band_instruments, by = "name"),
    tidylog::inner_join(band_members, band_instruments, by = "name"), iterations = 100
) %>%
    dplyr::select(expression, min, median, n_itr) %>%
    kable()
```

On a larger dataset (with many row duplications):

```{r message = FALSE}
N <- 1000
df1 <- tibble(x1 = rnorm(N), key = round(runif(N) * 10))
df2 <- tibble(x2 = rnorm(N), key = round(runif(N) * 10))

bench::mark(
    dplyr::inner_join(df1, df2, by = "key"),
    tidylog::inner_join(df1, df2, by = "key"), iterations = 100
) %>%
    dplyr::select(expression, min, median, n_itr) %>%
    kable()
```