Use patchwork to create a facet grid with strips

Question

I'm using a specific function that returns a ggplot in a grid-search algorithm, and I want to arrange the resulting plot in a plot grid, like facet_grid() would do.

I cannot use standard facets as each scenario can only be interpreted in a specific scale, so each plot has independent x/y limits.

Here is a reproducible example:

library(tidyverse)
library(patchwork)
get_plot = function(a, b) list(ggplot(iris, aes(Sepal.Length, Sepal.Width)) +
                                 geom_point() + ggtitle(paste0(a, "--", b)))
x = expand_grid(dist_fun = c("rnorm", "rexp"),
                assumption=c("linear", "square")) %>%
  rowwise() %>%
  mutate(plot=get_plot(dist_fun, assumption))
x
#> # A tibble: 4 × 3
#> # Rowwise: 
#>   dist_fun assumption plot  
#>   <chr>    <chr>      <list>
#> 1 rnorm    linear     <gg>  
#> 2 rnorm    square     <gg>  
#> 3 rexp     linear     <gg>  
#> 4 rexp     square     <gg>

#expected outcome, missing strips:
wrap_plots(x$plot)

^{Created on 2023-12-08 with reprex v2.0.2}

Is there a way to replace the ggtitle identification with facet-like strips?

Answer 1

Well, that's not about which function to use. (: To get a facet_grid like look using patchwork requires that you write yourself a custom function to manipulate your ggplot objects before passing them to wrap_plots. To this end I first added a column containing the numeric position in the patch. Then, using a custom function you could remove the axes and add the strips conditional on the position:

Note: While I love such patchwork exercises, IMHO the easier approach would be to return the data from your function or a list containing both data and the plot, then use rbind/bind_rows + facet_grid as already suggested in the comments.

library(tidyverse)
library(patchwork)

get_plot <- function(a, b) {
  list(ggplot(iris, aes(Sepal.Length, Sepal.Width)) +
    geom_point())
}
x <- expand_grid(
  dist_fun = c("rnorm", "rexp"),
  assumption = c("linear", "square")
) %>%
  mutate(plot = get_plot(dist_fun, assumption))

x |>
  mutate(pos = row_number()) |>
  pmap(
    \(...) {
      args <- list(...)
      col <- 1 + (args$pos + 1) %% 2
      row <- 1 + (args$pos - 1) %/% 2

      remove_x <- if (row == 1) {
        theme(
          axis.text.x = element_blank(),
          axis.ticks.x = element_blank(),
          axis.ticks.length.x = unit(0, "pt")
        )
      }
      remove_y <- if (col == 2) {
        theme(
          axis.text.y = element_blank(),
          axis.ticks.y = element_blank(),
          axis.ticks.length.y = unit(0, "pt")
        )
      }

      facet_y <- if (row == 1) {
        # Wrap in quotes to get e.g.
        # "linear" ~ . instead of linear ~ .
        paste0("\"", args$assumption, "\"")
      } else {
        "."
      }
      facet_x <- if (col == 2) {
        paste0("\"", args$dist_fun, "\"")
      } else {
        "."
      }

      layer_facet <- facet_grid(reformulate(facet_y, facet_x))

      args$plot +
        remove_x +
        remove_y +
        layer_facet
    }
  ) |>
  wrap_plots(ncol = 2) &
  theme(
    axis.title = element_blank()
  )

Answer 2

As others have said, it would be easiest to take your x object, strip out the data and the basic plot structure add facets. For example, the following function:

facetize <- function(x) {
  base_plot <- x$plot[[1]]
  x_nm <- names(x)
  base_plot$data <- x %>%
    mutate(plot_data = map(plot, `[[`, 'data')) %>%
    select(-plot) %>%
    unnest(plot_data)
  base_plot + 
    facet_grid(rows = vars(!!str2lang(x_nm[1])), 
               cols = vars(!!str2lang(x_nm[2])),
               scales = 'free')
}

Allows:

facetize(x)