One really cool feature from the ggplot2 package that I never really exploited enough was adding lists of layers to a plot. The fun thing about this was that I could pass a list of layers as an argument to a function and have them added to the plot. I could then get the desired appearance of the plot without necessarily returning the plot from the function (whether or not this is a good idea is another matter, but it was possible).
library(ggplot2)
x <- ggplot(mtcars,
aes(x = qsec,
y = mpg))
layers <- list(geom_point(),
geom_line(),
xlab("Quarter Mile Time"),
ylab("Fuel Efficiency"))
x + layers
Is there a way to do this with pipes? Something akin to:
#* Obviously isn't going to work
library(dplyr)
action <- list(group_by(am, gear),
summarise(mean = mean(mpg),
sd = sd(mpg)))
mtcars %>% action
To construct a sequence of magrittr steps, start with .
action = . %>% group_by(am, gear) %>% summarise(mean = mean(mpg), sd = sd(mpg))
Then it can be used as imagined in the OP:
mtcars %>% action
Like a list, we can subset to see each step:
action[[1]]
# function (.)
# group_by(., am, gear)
To review all steps, use functions(action) or just type the name:
action
# Functional sequence with the following components:
#
# 1. group_by(., am, gear)
# 2. summarise(., mean = mean(mpg), sd = sd(mpg))
#
# Use 'functions' to extract the individual functions.
Related
I have multiple graphs I'm generating with a data set. I preform many operations on the data (filtering rows, aggregating rows, calculations over columns, etc.) before passing on the result to ggplot(). I want to access the data I passed on to ggplot() in subsequent ggplot layers and facets so I can have more control over the resulting graph and to include some characteristics of the data in the plot itself, like for example the number of observations.
Here is a reproducible example:
library(tidyverse)
cars <- mtcars
# Normal scatter plot
cars %>%
filter(
# Many complicated operations
) %>%
group_by(
# More complicated operations
across()
) %>%
summarise(
# Even more complicated operations
n = n()
) %>%
ggplot(aes(x = mpg, y = qsec)) +
geom_point() +
# Join the dots but only if mpg < 20
geom_line(data = .data %>% filter(mpg < 20)) +
# Include the total number of observations in the graph
labs(caption = paste("N. obs =", NROW(.data)))
one could of course create a a separate data set before passing that onto ggplot and then reference that data set throughout (as in the example bellow). However, this is much more cumbersome as you need to save (and later remove) a data set for each graph and run two separate commands for just one graph.
I want to know if there is something that can be done that's more akin to the first example using .data (which obviously doesn't actually work).
library(tidyverse)
cars <- mtcars
tmp <- cars %>%
filter(
# Many complicated operations
) %>%
group_by(
# More complicated operations
across()
) %>%
summarise(
# Even more complicated operations
n = n()
)
tmp %>%
ggplot(aes(x = mpg, y = qsec)) +
geom_point() +
# Join the dots but only if mpg < 20
geom_line(data = tmp %>% filter(mpg < 20)) +
# Include the total number of observations in the graph
labs(caption = paste("N. obs =", NROW(tmp)))
Thanks for your help!
In the help page for each geom_ it helpfully gives a standard way:
A function will be called with a single argument, the plot data. The return value must be a data.frame, and will be used as the layer data. A function can be created from a formula (e.g. ~ head(.x, 10)).
For labs on the other hand you can use the . placeholders in piping, but you have to a) give the . as the data argument in the first place and b) wrap the whole thing in curly braces to recognise the later ..
So for example:
library(tidyverse)
cars <- mtcars
# Normal scatter plot
cars %>%
filter() %>%
group_by(across()) %>%
summarise(n = n()) %>%
{
ggplot(., aes(x = mpg, y = qsec)) +
geom_point() +
geom_line(data = ~ filter(.x, mpg < 20)) +
labs(caption = paste("N. obs =", NROW(.)))
}
Or if you don't like the purrr formula syntax, then the flashy new R anonymous functions work too:
geom_line(data = \(x) filter(x, mpg < 20)) +
Unfortunately the labs function doesn't seem to have an explicit way of testing whether data is shuffling invisibly through the ggplot stack as by-and-large it usually can get on with its job without touching the main data. These are some ways around this.
#Preparing the data and loading packages
library(modelsummary);library(tidyverse);library(gt)
as_tibble(mtcars)
df <- mtcars %>% mutate(cyl_ = factor(cyl)) %>%
dplyr::select(cyl_, mpg, vs, am, hp, wt)
#Gets table of descriptive statistics about different subsets of the data
print(t1 <- datasummary_balance(~cyl_,
data = df,
output = "gt"))
#This hides the "Std. Dev." columns
t1 %>% cols_hide(c(3,5,7))
#Now I want to hide the "Mean" column labels, but I want to keep the "cyl_" value column labels. Any ideas how?
I want something like this:
Using the gt package, you can pipe your table to tab_options(column_labels.hidden = TRUE) to remove column labels. Unfortunately, this will remove both levels: the column headers, and the spanning labels that include the cyl info you want to keep.
Note that datasummary_balance() produces a highly customized table which is intended to be used as a ready-made output. In cases like these, it might be easier to just build the custom table you want using datasummary() instead of trying to customize datasummary_balance() (square peg, round hole, etc). For example:
library(modelsummary)
library(tidyverse)
df <- mtcars %>%
select(cyl, mpg, vs, am, hp, wt) %>%
mutate(cyl = factor(sprintf("%s (N = %s)", cyl, n()))) %>%
as.data.frame() # The `All()` function does not accept tibbles
datasummary(
All(df) ~ Mean * cyl,
data = df,
output = "gt")
I'm looking for a way to apply a function to either specified labels, or to all labels that are included in the plot. The goal is to have neat human readable labels that derive from the default labels, without having to specify each.
To demonstrate what I am looking for in terms of the input variable names and the output, I am including an example based on the starwars data set, that uses the versatile snakecase::to_sentence_case() function, but this could apply to any function, including ones that expand short variable names in pre-determined ways:
library(tidyverse)
library(snakecase)
starwars %>%
filter(mass < 1000) %>%
mutate(species = species %>% fct_infreq %>% fct_lump(5) %>% fct_explicit_na) %>%
ggplot(aes(height, mass, color=species, size=birth_year)) +
geom_point() +
labs(
x = to_sentence_case("height"),
y = to_sentence_case("mass"),
color = to_sentence_case("species"),
size = to_sentence_case("birth_year")
)
Which produces the following graph:
The graph is the desired output, but requires that each of the labels be specified by hand, increasing the possibility of error if the variables are later changed. Note that if I had not specified the labels, all the labels would have been applied automatically, but with the variable names instead of the prettier versions.
This issue seems to be somewhat related to what the labeller() function is intended for, but it seems that it only applies to facetting. Another related issue is raised in this question. However, both of these seem to apply only to values contained within the data, not to the variable names that are being used in the plot, which is what I am looking for.
The very helpful answer by #z-lin demonstrated to me a simple way to do this by simply modifying the plot object before printing.
The intended result can be achieved with the help of gg_apply_labs(), a short function that will apply an arbitrary string processing function to the $labels of a plot object. The resulting code should be a self-contained illustration of this approach:
# Packages
library(tidyverse)
library(snakecase)
# This applies fun to each label present in the plot object
#
# fun should accept and return character vectors, it can either be a simple
# prettyfying function or it can perform more complex lookup to replace
# variable names with variable labels
gg_apply_labs <- function(p, fun) {
p$labels <- lapply(p$labels, fun)
p
}
# This gives the intended result
# Note: The plot is assigned to a named variable before piping to apply_labs()
p <- starwars %>%
filter(mass < 1000) %>%
mutate(species = species %>% fct_infreq %>% fct_lump(5) %>% fct_explicit_na) %>%
ggplot(aes(height, mass, color=species, size=birth_year)) +
geom_point()
p %>% gg_apply_labs(to_sentence_case)
# This also gives the intended result, in a single pipeline
# Note: It is important to put in the extra parentheses!
(starwars %>%
filter(mass < 1000) %>%
mutate(species = species %>% fct_infreq %>% fct_lump(5) %>% fct_explicit_na) %>%
ggplot(aes(height, mass, color=species, size=birth_year)) +
geom_point()) %>%
gg_apply_labs(to_sentence_case)
# This DOES NOT give the intended result
# Note: The issue is probably order precedence
starwars %>%
filter(mass < 1000) %>%
mutate(species = species %>% fct_infreq %>% fct_lump(5) %>% fct_explicit_na) %>%
ggplot(aes(height, mass, color=species, size=birth_year)) +
geom_point() %>%
gg_apply_labs(to_sentence_case)
A simple solution is to pipe through rename_all (or rename_if if you want more control) before plotting:
library(tidyverse)
library(snakecase)
starwars %>%
filter(mass<1000) %>%
mutate(species=species %>% fct_infreq %>% fct_lump(5) %>% fct_explicit_na) %>%
rename_all(to_sentence_case) %>%
#rename_if(is.character, to_sentence_case) %>%
ggplot(aes(Height, Mass, color=Species, size=`Birth year`)) +
geom_point()
#> Warning: Removed 23 rows containing missing values (geom_point).
Created on 2019-11-25 by the reprex package (v0.3.0)
Note, though, that the variables given to aes in ggplot in this case must be modified to match the modified sentence case variable names.
You can modify a ggplot object's appearance at the point of printing / plotting it, without affecting the original plot object, using trace:
trace(what = ggplot2:::ggplot_build.ggplot,
tracer = quote(plot$labels <- lapply(plot$labels,
<whatever string function you desire>)))
This will change the appearance of all existing / new ggplot objects you wish to plot / save, until you turn off the trace via either untrace(...) or tracingState(on = FALSE).
Illustration
Create a normal plot with default labels in lower case:
library(tidyverse)
p <- starwars %>%
filter(mass < 1000) %>%
mutate(species=species %>% fct_infreq %>% fct_lump(5) %>% fct_explicit_na) %>%
ggplot(aes(height, mass, color=species, size=birth_year)) +
geom_point() +
theme_bw()
p # if we print the plot now, all labels will be lower-case
Apply a function to modify the appearance of all labels:
trace(what = ggplot2:::ggplot_build.ggplot,
tracer = quote(plot$labels <- lapply(plot$labels,
snakecase::to_sentence_case)))
p # all labels will be in sentence case
trace(what = ggplot2:::ggplot_build.ggplot,
tracer = quote(plot$labels <- lapply(plot$labels,
snakecase::to_screaming_snake_case)))
p # all labels will be in upper case
trace(what = ggplot2:::ggplot_build.ggplot,
tracer = quote(plot$labels <- lapply(plot$labels,
snakecase::to_random_case)))
p # all letters in all labels may be in upper / lower case randomly
# (exact order can change every time we print the plot again, unless we set the same
# random seed for reproducibility)
trace(what = ggplot2:::ggplot_build.ggplot,
tracer = quote(plot$labels <- lapply(plot$labels,
function(x) paste("!!!", x, "$$$"))))
p # all labels now have "!!!" in front & "$$$" behind (this is a demonstration for
# an arbitrary user-defined function, not a demonstration of good taste in labels)
Toggle between applying & not applying the function:
tracingState(on = FALSE)
p # back to sanity, temporarily
tracingState(on = TRUE)
p # plot labels are affected by the function again
untrace(ggplot2:::ggplot_build.ggplot)
p # back to sanity, permanently
I'm using do() to fit a model to grouped data, and then I want to plot the fit for each group. In plyr, I guess I would use d_ply(). In dplyr, I'm trying either do() or summarise() using a function that makes the plot as a side effect.
I'm getting different results depending on whether I use do() or summarise(), and I'm not sure why. Specifically it seems like summarise() isn't operating on each row correctly.
Here's my example:
require(nycflights13)
require(mgcv)
# fit a gam to the flights grouped by dest (from ?do)
by_dest <- flights %>% group_by(dest) %>% filter(n() > 100)
models = by_dest %>% do(smooth = gam(arr_delay ~ s(dep_time) + month, data = .))
# print the first 4 rows, the dest is ABQ, ACK, ALB, ATL
models %>% slice(1:4)
# make a function to plot the models, titled by dest
plot.w.title = function(title, gam.model){
plot.gam(gam.model, main=title)
return(1)
}
# This code makes plots with the wrong titles, for example ATL is listed twice:
models %>%
slice(1:4) %>%
rowwise %>%
summarise(useless.column = plot.w.title(dest, smooth)) # for plot side effect
# this code gives me the correct titles...why the difference?
models %>%
slice(1:4) %>%
rowwise %>%
do(useless.column = plot.w.title(.$dest, .$smooth))
The summarise() method will work if you modify the function by applying unique() to the title:
plot.w.title = function(title, gam.model){
plot.gam(gam.model, main=unique(title))
return(1)
}
Trying to replicate the ggplot function position="fill" in ggvis. I use this handy function all the time in the presentation of results. Reproducible example successfully performed in ggplot2 + the ggvis code. Can it be done using the scale_numeric function?
library(ggplot2)
p <- ggplot(mtcars, aes(x=factor(cyl), fill=factor(vs)))
p+geom_bar()
p+geom_bar(position="fill")
library(ggvis)
q <- mtcars %>%
ggvis(~factor(cyl), fill = ~factor(vs))%>%
layer_bars()
# Something like this?
q %>% scale_numeric("y", domain = c(0,1))
I think that to do this sort of thing with ggvis you have to do the heavy data reshaping lifting before sending it to ggvis. ggplot2's geom_bar handily does a lot of calculations (counting things up, weighting them, etc) for you that you need to do explicitly yourself in ggvis. So try something like the below (there may be more elegant ways):
mtcars %>%
mutate(cyl=factor(cyl), vs=as.factor(vs)) %>%
group_by(cyl, vs) %>%
summarise(count=length(mpg)) %>%
group_by(cyl) %>%
mutate(proportion = count / sum(count)) %>%
ggvis(x= ~cyl, y = ~proportion, fill = ~vs) %>%
layer_bars()