ggplot2: facet_wrap failure for multi-layered plots with two variables - r

Question

I am trying to create a multi-faceted plot with free scaling using ggplot2. By design, facet_grid, cannot achieve what I need. And facet_wrap fails with a cryptic error. Could you please tell me, do you have any suggestions on how to fix the error? A reproducible example is given below.
Let's create sample data:
require(tidyverse)
require(modelr)
d1 <- tibble(
x = 1:100,
y = 1:100 + rnorm(10),
z = y ^ 2,
dataset_name = "d1"
)
d2 <- tibble(
x = 1:1000,
y = 1:1000 + rnorm(10),
z = y ^ 2,
dataset_name = "d2"
)
#these data will be used for the 1st layer
actuals <- bind_rows(d1, d2)
#these data will be used for the 2nd layer
predictions <- bind_rows(
d1 %>% gather_predictions(
"m1" = lm(y ~ x, data = d1),
"m2" = lm(y ~ x + z, data = d1),
.pred = "y"
),
d2 %>% gather_predictions(
"m1" = lm(y ~ x, data = d2),
"m2" = lm(y ~ x + z, data = d2),
.pred = "y"
)
)
facet_grid generated the required graphs:
)
But it cannot (by design) scale the x-axis:
ggplot(actuals, aes(x, y)) +
geom_point() +
geom_line(data = predictions, colour = "red") +
facet_grid(dataset_name ~ model, scales = "free")
If I want to plot the data only for one dataset (namely, predictions), it works as expected and I get 4 facets:
ggplot(predictions, aes(x, y)) +
geom_point() +
facet_wrap( ~ model + dataset_name, scales = "free")
However, if I try to combine actuals and predictions as follows:
ggplot(actuals, aes(x, y)) +
geom_point() +
geom_line(data = predictions, colour = "red") +
facet_wrap( ~ model + dataset_name, scales = "free")
Then things fall apart with the following error: Error in gList(list(x = 0.5, y = 0.5, width = 1, height = 1, just = "centre", : only 'grobs' allowed in "gList"

Answer 1

Try making a single variable with the interaction of model and dataset_name.
# these two blocks of code are equivalent
library(magrittr)
predictions %<>% mutate(mod_dn = interaction(model, dataset_name))
and
predictions <- predictions %>%
mutate(mod_dn = interaction(model, dataset_name))
Now, this poses a problem for facet_wrap, since mod_dn does not exist there. So we need to merge the two datasets together. Using tidyverse, we can do this with left_join, but we need to be careful about what we join by, and then adjust the ggplot call accordingly:
all_data <- left_join(
actuals,
predictions,
by = c("x", "dataset_name"),
suffix = c(".actual", ".pred")
)
all_data %>%
ggplot(aes(x, y.actual)) +
geom_point() +
geom_line(aes(y = y.pred), colour = "red") +
facet_wrap( ~ mod_dn, scales = "free") +
labs(y = "y")

Answer 2

Thank you all for the help! It seems that a more straightforward solution (even though it will alter the original question), would be to tweak the predictions as follows:
predictions <- bind_rows(
d1 %>% gather_predictions(
"m1" = lm(y ~ x, data = d1),
"m2" = lm(y ~ x + z, data = d1),
.pred = "y.pred"
),
d2 %>% gather_predictions(
"m1" = lm(y ~ x, data = d2),
"m2" = lm(y ~ x + z, data = d2),
.pred = "y.pred"
)
)
Then we can do the plotting without resorting to joins:
ggplot(predictions, aes(x, y)) +
geom_point() +
geom_line(aes(x, y.pred), colour = "red") +
facet_wrap( ~ model + dataset_name, scales = "free")
This renders the desired plot.