Get centroids — get_centroids • joeyr

Runs a summarizing function for each specified column, for each specified group. This is intended to be used to plot centroids in ellipses in ggplot2 without having to create a new object or have a lot of in-line code. See examples below.

get_centroids(df, .cols, ..., .fns = median)

Arguments

df: a dataframe.
.cols: columns that should be summarized. For sociophonetic data, this is usually the names of your vowel columns, e.g. c(F1, F2). This literally is just passed into an `across` function within `summarize`.
...: grouping variables. For sociophonetic data, this might be speaker and allophone or something. This is just passed into `group_by`.
.fns: one or more names of functions. By default, median. This is passed into `across`.

Value

an ungrouped dataframe

Note

Okay technically this function name is a misnomer because we're not truly getting centroids in a mathematical sense. But that's what I think of when I run this so that's what we're going with.

Examples

library(tidyverse)
df <- joeysvowels::idahoans

# Basic usage as a summarizing function
df %>%
  get_centroids(c(F1, F2), vowel)
#> # A tibble: 11 × 3
#>    vowel    F1    F2
#>    <chr> <dbl> <dbl>
#>  1 AA     746. 1288.
#>  2 AE     712. 1701.
#>  3 AH     673. 1460.
#>  4 AO     580. 1001.
#>  5 EH     621. 1750.
#>  6 EY     454. 2292.
#>  7 IH     489. 1957.
#>  8 IY     372. 2364.
#>  9 OW     540. 1176.
#> 10 UH     488. 1470.
#> 11 UW     399. 1693.

# Within a ggplot2 block. Note that you do have to start the data argument with the dot and pipe it into get_centroids, rather than incorporating it in (i.e. get_centroids(., vowel)). Not sure why but this appears to be a contraint imposed by ggplot2.
ggplot(df, aes(F2, F1, color = vowel)) +
  stat_ellipse(level = 0.67) +
  geom_text(data = . %>% get_centroids(c(F1, F2), vowel), aes(label = vowel)) +
  scale_x_reverse() +
  scale_y_reverse() +
  theme(legend.position = "none")


# You can add multiple groups to the code too.
ggplot(df, aes(F2, F1, color = vowel)) +
  stat_ellipse(level = 0.67) +
  geom_text(data = . %>% get_centroids(c(F1, F2), speaker, vowel), aes(label = vowel)) +
  scale_x_reverse() +
  scale_y_reverse() +
  facet_wrap(~speaker, scales = "free") +
  theme(legend.position = "none")


# Like any use of group_by(), additional, perhaps redundant columns may be specified for the purpose of "passing them through." In this example, adding tense_lax doesn't change the calculations, but it's useful for this plot. Additionally, this block of code highlights one strength of get_centroids, and that is that I can pass in a modified dataframe directly to ggplot and then modify it even further to get the labels, without needing to create any new objects.
df %>%
  mutate(tense_lax = fct_collapse(vowel,
                                  "tense" = c("IY", "EY", "AO", "OW", "UW"),
                                  "lax"   = c("IH", "EH", "AE", "AA", "AH", "UH"))) %>%
  ggplot(aes(F2, F1, color = tense_lax, group = vowel)) +
  stat_ellipse(level = 0.67) +
  geom_text(data = . %>% get_centroids(c(F1, F2), speaker, tense_lax, vowel),
            aes(label = vowel)) +
  scale_x_reverse() +
  scale_y_reverse() +
  facet_wrap(~speaker, scales = "free") +
  theme(legend.position = "none")


# For column selection, any tidyselect output works, such as matches().
df %>%
  get_centroids(matches("F\\d"), speaker, vowel)
#> # A tibble: 110 × 6
#>    speaker vowel    F1    F2    F3    F4
#>    <fct>   <chr> <dbl> <dbl> <dbl> <dbl>
#>  1 01      AA     722. 1381. 1974. 3197.
#>  2 01      AE     675. 2058. 2569. 3389.
#>  3 01      AH     650. 1814. 2389. 3639.
#>  4 01      AO     536.  857. 2329. 3606.
#>  5 01      EH     653. 1884. 2409. 3621.
#>  6 01      EY     498. 2281. 2905. 3735.
#>  7 01      IH     517. 2154. 2788. 3783.
#>  8 01      IY     352. 2784. 3015. 4152.
#>  9 01      OW     566. 1405. 2562. 3713.
#> 10 01      UH     533. 1690. 2524. 3745.
#> # … with 100 more rows

# For functions, you can add more than one. Just wrap them up into c().
df %>%
  get_centroids(c(F1, F2), .fns = c(median, mean), speaker, vowel)
#> # A tibble: 110 × 6
#>    speaker vowel  F1_1  F1_2  F2_1  F2_2
#>    <fct>   <chr> <dbl> <dbl> <dbl> <dbl>
#>  1 01      AA     722.  731. 1381. 1405.
#>  2 01      AE     675.  644. 2058. 2033.
#>  3 01      AH     650.  654. 1814. 1805.
#>  4 01      AO     536.  582.  857.  922.
#>  5 01      EH     653.  635. 1884. 1937.
#>  6 01      EY     498.  483. 2281. 2265.
#>  7 01      IH     517.  513. 2154. 2147.
#>  8 01      IY     352.  350. 2784. 2713.
#>  9 01      OW     566.  564. 1405. 1353.
#> 10 01      UH     533.  530. 1690. 1615.
#> # … with 100 more rows

# However, unless they are named, they won't be useful.
df %>%
  get_centroids(c(F1, F2), .fns = c(`med` = median, `average` = mean), speaker, vowel)
#> # A tibble: 110 × 6
#>    speaker vowel F1_med F1_average F2_med F2_average
#>    <fct>   <chr>  <dbl>      <dbl>  <dbl>      <dbl>
#>  1 01      AA      722.       731.  1381.      1405.
#>  2 01      AE      675.       644.  2058.      2033.
#>  3 01      AH      650.       654.  1814.      1805.
#>  4 01      AO      536.       582.   857.       922.
#>  5 01      EH      653.       635.  1884.      1937.
#>  6 01      EY      498.       483.  2281.      2265.
#>  7 01      IH      517.       513.  2154.      2147.
#>  8 01      IY      352.       350.  2784.      2713.
#>  9 01      OW      566.       564.  1405.      1353.
#> 10 01      UH      533.       530.  1690.      1615.
#> # … with 100 more rows