Share
Embed
Transcript
120 slides
-
Data analysis programming Production programming • Structure specific tasks data
manipulation, data cleaning, plotting, ... • Interactivity and iteration • Reproducibility by few users • Structure repeated computations functional programming, generic programming, metaprogramming, ... • Flexibility and robustness • Reusability by many users r-lib tidyverse -
Programming in the tidyverse • Tidyverse optimised for interactive analyses
• Moving towards code reusability • How to program with the tidyverse • Demystifying tidy evaluation -
What is the tidyverse • It is a set of
packages (dplyr, tidyr, purrr, ggplot2, ...) • Website: https://tidyverse.org • Meta-package: library(tidyverse) -
What is the tidyverse • It is a set of
principles • Human centered Computers + people • Consistent Reuse small set of ideas • Composable Solve larger problems • Inclusive Diverse community https://principles.tidyverse.org -
What is the tidyverse • It is a set of
principles • Human centered Computers + people • Consistent Reuse small set of ideas • Composable Solve larger problems • Inclusive Diverse community https://principles.tidyverse.org -
Human centered Tidyverse packages ⟶ • Domain oriented • Language-like
interface • Data is the important scope -
• Domain oriented • Language-like interface • Data is the
important scope Set of verbs for data manipulation • select() • filter() • arrange() • mutate() • group_by() • summarise() -
flights # A tibble: 336,776 x 19 year month day
dep_time sched_dep_time dep_delay arr_time <int> <int> <int> <int> <int> <dbl> <int> 1 2013 1 1 517 515 2 830 2 2013 1 1 533 529 4 850 3 2013 1 1 542 540 2 923 4 2013 1 1 544 545 -1 1004 # … with 336,772 more rows, and 12 more variables: sched_arr_time <int>, # arr_delay <dbl>, carrier <chr>, flight <int>, tailnum <chr>, # origin <chr>, dest <chr>, air_time <dbl>, distance <dbl>, hour <dbl>, … -
flights %>% filter(month == 10, day == 10) # A
tibble: 687 x 19 year month day dep_time sched_dep_time dep_delay arr_time <int> <int> <int> <int> <int> <dbl> <int> 1 2013 10 5 453 500 -7 624 2 2013 10 5 525 515 10 747 3 2013 10 5 541 545 -4 827 4 2013 10 5 542 545 -3 813 # … with 683 more rows, and 12 more variables: sched_arr_time <int>, # arr_delay <dbl>, carrier <chr>, flight <int>, tailnum <chr>, # origin <chr>, dest <chr>, air_time <dbl>, distance <dbl>, hour <dbl>, … -
flights %>% arrange(desc(month), desc(day)) # A tibble: 336,776 x 19
year month day dep_time sched_dep_time dep_delay arr_time <int> <int> <int> <int> <int> <dbl> <int> 1 2013 12 31 13 2359 14 439 2 2013 12 31 18 2359 19 449 3 2013 12 31 26 2245 101 129 4 2013 12 31 459 500 -1 655 # … with 336,772 more rows, and 12 more variables: sched_arr_time <int>, # arr_delay <dbl>, carrier <chr>, flight <int>, tailnum <chr>, # origin <chr>, dest <chr>, air_time <dbl>, distance <dbl>, hour <dbl>, # minute <dbl>, time_hour <dttm> -
flights %>% select(year, month, day) # A tibble: 336,776 x
3 year month day <int> <int> <int> 1 2013 1 1 2 2013 1 1 3 2013 1 1 4 2013 1 1 # … with 336,772 more rows -
flights %>% select(year:day) # A tibble: 336,776 x 3 year
month day <int> <int> <int> 1 2013 1 1 2 2013 1 1 3 2013 1 1 4 2013 1 1 # … with 336,772 more rows Equivalent to select(1:3) -
flights %>% select(ends_with("_time")) # A tibble: 336,776 x 5 dep_time
sched_dep_time arr_time sched_arr_time air_time <int> <int> <int> <int> <dbl> 1 517 515 830 819 227 2 533 529 850 830 227 3 542 540 923 850 160 4 544 545 1004 1022 183 # … with 336,772 more rows -
flights %>% mutate( gain = arr_delay - dep_delay, gain_per_hour =
gain / (air_time / 60) ) # A tibble: 336,776 x 21 year month day dep_time sched_dep_time dep_delay arr_time <int> <int> <int> <int> <int> <dbl> <int> 1 2013 1 1 517 515 2 830 2 2013 1 1 533 529 4 850 3 2013 1 1 542 540 2 923 4 2013 1 1 544 545 -1 1004 # … with 336,772 more rows, and 14 more variables: sched_arr_time <int>, # arr_delay <dbl>, carrier <chr>, flight <int>, tailnum <chr>, # origin <chr>, dest <chr>, air_time <dbl>, distance <dbl>, hour <dbl>, … -
flights %>% transmute( gain = arr_delay - dep_delay, gain_per_hour =
gain / (air_time / 60) ) # A tibble: 336,776 x 2 gain gain_per_hour <dbl> <dbl> 1 9 2.38 2 16 4.23 3 31 11.6 4 -17 -5.57 # … with 336,772 more rows • mutate() adds columns •transmute() creates a new tibble -
flights %>% group_by(month) %>% summarise(avg_delay = mean(arr_delay, na.rm = TRUE))
# A tibble: 12 x 2 month n <int> <dbl> 1 1 6.13 2 2 5.61 3 3 5.81 4 4 11.2 5 5 3.52 6 6 16.5 7 7 16.7 8 8 6.04 9 9 -4.02 10 10 -0.167 11 11 0.461 12 12 14.9 • group_by() only affects future computations •summarise() makes one summary per level -
starwars[starwars$height < 200 & starwars$gender == "male", ] starwars %>%
filter( height < 200, gender == "male" ) • Domain oriented • Language-like interface • Data is the important scope Data masking -
Data masking data %>% fill(year) %>% spread(key, count) starwars %>%
ggplot(aes(height, mass)) + geom_point() + facet_wrap(vars(hair_color)) starwars %>% filter( height < 200, gender == "male" ) -
Data masking starwars %>% base::subset(height < 150, name:mass) %>% base::transform(height
= height / 100) starwars %>% stats::lm(formula = mass ~ height) In base R too! • Inspiration for dplyr • By R core member Peter Dalgaard -
Data masking • Unique feature of R • Great for
data analysis • Focus on the task not the subsetting • Programming is a bit more involved Tidy eval framework -
Tidy eval • Powers data masking from the rlang package
• Flexible and robust programming • Strange syntax: !! and !!!, enquo(), etc • Requires learning new concepts -
Demystifying tidy eval 1. Why tidy evaluation? 2. Do you
actually need it? 3. Can it be used by laypeople? -
Why Tidy Eval starwars[starwars$height < 200 & starwars$gender == "male",
] starwars %>% filter( height < 200, gender == "male" ) Change the context of computation -
Why Tidy Eval starwars %>% filter( height < 200, gender
== "male" ) <SQL> SELECT * FROM `starwars` WHERE ((`height` < 200.0) AND (`gender` = 'male')) Change the context of computation -
Why Tidy Eval ⟶ Need to delay computations list( height
< 200, gender == "male" ) Error: object 'height' not found starwars %>% filter( height < 200, gender == "male" ) -
Why Tidy Eval How it works • Delay computations by
quoting • Change the context and resume computation starwars %>% filter( height < 200, gender == "male" ) -
Quoted code is like a blueprint vars( height < 200,
gender == "male" ) [[1]] <quosure> expr: ^height < 200 env: global [[2]] <quosure> expr: ^gender == "male" env: global • vars() is a fundamental quoting function • Returns blueprints of delayed computations -
Quoted code is like a blueprint Flip side: Harder to
reuse with different inputs • Loops • Functions -
columns <- c("hair_color", "skin_color") out <- rep(list(NULL), 2) for (i
in seq_along(columns)) { out[[i]] <- starwars %>% summarise(avg = mean(columns[[i]], na.rm = TRUE)) } -
columns <- c("hair_color", "skin_color") out <- rep(list(NULL), 2) for (i
in seq_along(columns)) { out[[i]] <- starwars %>% summarise(avg = mean(columns[[i]], na.rm = TRUE)) } out[[1]] # A tibble: 1 x 1 avg <dbl> 1 NA -
columns <- c("hair_color", "skin_color") out <- rep(list(NULL), 2) for (i
in seq_along(columns)) { out[[i]] <- starwars %>% summarise(avg = mean(columns[[i]], na.rm = TRUE)) } out[[1]] # A tibble: 1 x 1 avg <dbl> 1 NA mean("hair_color", na.rm = TRUE) [1] NA Warning message: argument is not numeric or logical: returning NA -
average <- function(data, x) { data %>% summarise(avg = mean(x,
na.rm = TRUE)) } -
average <- function(data, x) { data %>% summarise(avg = mean(x,
na.rm = TRUE)) } average(starwars, "hair_color") # A tibble: 1 x 1 avg <dbl> 1 NA Warning message: argument is not numeric or logical: returning NA -
average <- function(data, x) { data %>% summarise(avg = mean(x,
na.rm = TRUE)) } average(starwars, hair_color) Error: object 'hair_color' not found • Data masking is not transitive • x is masked instead of hair_color -
Quoted code is like a blueprint Programming requires modifying the
blueprint • !! and !!! are surgery operators for blueprints • Need blueprint material: sym(), enquo(), ... Metaprogramming skills -
But before we get there... Do you need tidy eval?
• Fixed column names • Columnwise mapping -
• No need for tidy eval when column names are
fixed! • Trivial to implement Fixed column names -
data %>% transmute(bmi = mass / height^2) compute_bmi <- function(data)
{ data %>% transmute(bmi = mass / height^2) } compute_bmi <- function(data) { if (!all(c("mass", "height") %in% names(data))) { stop("`data` must contain `mass` and `height` columns") } data %>% transmute(bmi = mass / height^2) } 1. Repeated code -
data %>% transmute(bmi = mass / height^2) compute_bmi <- function(data)
{ data %>% transmute(bmi = mass / height^2) } compute_bmi <- function(data) { if (!all(c("mass", "height") %in% names(data))) { stop("`data` must contain `mass` and `height` columns") } data %>% transmute(bmi = mass / height^2) } 2. Wrap pipeline -
data %>% transmute(bmi = mass / height^2) 3. Check inputs
compute_bmi <- function(data) { data %>% transmute(bmi = mass / height^2) } compute_bmi <- function(data) { if (!all(c("mass", "height") %in% names(data))) { stop("`data` must contain `mass` and `height` columns") } data %>% transmute(bmi = mass / height^2) } -
compute_bmi <- function(data) { if (!all(c("mass", "height") %in% names(data))) {
stop("`data` must contain `mass` and `height` columns") } mean_height <- round(mean(data$height, na.rm = TRUE), 1) if (mean_height > 3) { warning(glue::glue( "Average height is { mean_height }, is it scaled in meters?" )) } data %>% transmute(bmi = mass / height^2) } 4. Check inputs! -
starwars %>% compute_bmi() # A tibble: 87 x 1 bmi
<dbl> 1 0.00260 2 0.00269 3 0.00347 4 0.00333 # … with 83 more rows Warning message: Average height is 174.4, is it scaled in meters? -
starwars %>% mutate(height = height / 100) %>% compute_bmi() #
A tibble: 87 x 1 bmi <dbl> 1 26.0 2 26.9 3 34.7 4 33.3 # … with 83 more rows -
• For specific tasks where fixed names make sense •
Callers must ensure existence of these columns • Input checking is important • Domain logic may have greater payoff Fixed column names -
Columnwise mapping • Repeat operations across columns by mapping •
Important notion in R (apply family) • The purrr package is all about mapping! -
map() loops a function automatically map(data, n_distinct) $group [1] 3
$value [1] 5 n_distinct(data$group) [1] 3 n_distinct(data$value) [1] 5 -
Scoped variants for dplyr verbs • Map functions over a
selection of columns • _all suffix ⟶ Map over all columns • _if suffix ⟶ Map over columns selected by a predicate • _at suffix ⟶ Map over a custom selection • Full dplyr features, including groups support Columnwise mapping -
Columnwise mapping No data masking • Take objects not blueprints
• Easy to program with -
Columnwise mapping mutate_all() mutate_at() mutate_if() summarise_all() summarise_at() summarise_if() group_by_all() group_by_at()
group_by_if() filter_all() filter_at() filter_if() -
mtcars %>% mutate_all(function(x) x / sd(x)) mtcars %>% mutate_all(~ .
/ sd(.)) # A tibble: 32 x 11 mpg cyl disp hp drat wt qsec vs am gear carb <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> 1 3.48 3.36 1.29 1.60 7.29 2.68 9.21 0 2.00 5.42 2.48 2 3.48 3.36 1.29 1.60 7.29 2.94 9.52 0 2.00 5.42 2.48 3 3.78 2.24 0.871 1.36 7.20 2.37 10.4 1.98 2.00 5.42 0.619 4 3.55 3.36 2.08 1.60 5.76 3.29 10.9 1.98 0 4.07 0.619 # … with 28 more rows Mapping a function over all columns Supports purrr formulas for anonymous functions -
iris %>% mutate_if(is.numeric, ~ . / sd(.)) # A tibble:
150 x 5 Sepal.Length Sepal.Width Petal.Length Petal.Width Species <dbl> <dbl> <dbl> <dbl> <fct> 1 6.16 8.03 0.793 0.262 setosa 2 5.92 6.88 0.793 0.262 setosa 3 5.68 7.34 0.736 0.262 setosa 4 5.56 7.11 0.850 0.262 setosa # … with 146 more rows Mapping a function over predicate selection This function determines which columns are changed -
iris %>% mutate_at(1:4, ~ . / sd(.)) # A tibble:
150 x 5 Sepal.Length Sepal.Width Petal.Length Petal.Width Species <dbl> <dbl> <dbl> <dbl> <fct> 1 6.16 8.03 0.793 0.262 setosa 2 5.92 6.88 0.793 0.262 setosa 3 5.68 7.34 0.736 0.262 setosa 4 5.56 7.11 0.850 0.262 setosa # … with 146 more rows Mapping a function over custom selection Numeric vector of column positions -
nms <- c("Sepal.Length", "Sepal.Width", "Petal.Length", "Petal.Width") iris %>% mutate_at(nms, ~
. / sd(.)) # A tibble: 150 x 5 Sepal.Length Sepal.Width Petal.Length Petal.Width Species <dbl> <dbl> <dbl> <dbl> <fct> 1 6.16 8.03 0.793 0.262 setosa 2 5.92 6.88 0.793 0.262 setosa 3 5.68 7.34 0.736 0.262 setosa 4 5.56 7.11 0.850 0.262 setosa # … with 146 more rows Mapping a function over custom selection Character vector of column names -
iris %>% mutate_at(vars(Sepal.Length, Sepal.Width), ~ . / sd(.)) iris %>%
mutate_at(vars(starts_with("Sepal")), ~ . / sd(.)) # A tibble: 150 x 5 Sepal.Length Sepal.Width Petal.Length Petal.Width Species <dbl> <dbl> <dbl> <dbl> <fct> 1 6.16 8.03 1.4 0.2 setosa 2 5.92 6.88 1.4 0.2 setosa 3 5.68 7.34 1.3 0.2 setosa 4 5.56 7.11 1.5 0.2 setosa # … with 146 more rows Mapping a function over custom selection Pass selection helpers with vars() -
iris %>% summarise_if(is.numeric, mean) # A tibble: 1 x 4
Sepal.Length Sepal.Width Petal.Length Petal.Width <dbl> <dbl> <dbl> <dbl> 1 5.84 3.06 3.76 1.20 Consistent behaviour across variants -
iris %>% group_by(Species) %>% summarise_all(mean) # A tibble: 3 x
5 Species Sepal.Length Sepal.Width Petal.Length Petal.Width <fct> <dbl> <dbl> <dbl> <dbl> 1 setosa 5.01 3.43 1.46 0.246 2 versicolor 5.94 2.77 4.26 1.33 3 virginica 6.59 2.97 5.55 2.03 Consistent behaviour across variants -
iris %>% group_by_if(is.factor) %>% summarise_at(vars(starts_with("Sepal")), mean) # A tibble: 3
x 3 Species Sepal.Length Sepal.Width <fct> <dbl> <dbl> 1 setosa 5.01 3.43 2 versicolor 5.94 2.77 3 virginica 6.59 2.97 Consistent behaviour across variants -
Columnwise mapping • Scoped variants can be incredibly useful •
Reuse skills from purrr and apply functions • No tidy eval needed for looping -
1. Pass the dots 2. Subset .data 3. Interpolate Tidy
eval, the easy parts -
1. Pass the dots 2. Subset .data 3. Interpolate Tidy
eval, the easy parts -
Passing the dots Easiest way to create a tidy eval
function! starwars %>% group_by(gender) %>% summarise(n = n()) -
my_count_by <- function(data, ...) { data %>% group_by(...) %>% summarise(n
= n()) } Easiest way to create a tidy eval function! Passing the dots -
starwars %>% my_count_by(gender) # A tibble: 5 x 2 gender
n <chr> <int> 1 NA 3 2 female 19 3 hermaphrodite 1 4 male 62 5 none 2 my_count_by <- function(data, ...) { data %>% group_by(...) %>% summarise(n = n()) } Passing the dots -
Recipient of dots takes care of everything! • No need
to quote / delay blueprints • Properties of the function are inherited Passing the dots -
Two flavours starwars %>% mutate(birth_year - 100) starwars %>% group_by(birth_year)
starwars %>% select(birth_year) starwars %>% filter(birth_year < 50) One of these things is not like the other things! -
Two flavours starwars %>% mutate(birth_year - 100) starwars %>% group_by(birth_year)
starwars %>% select(birth_year) starwars %>% filter(birth_year < 50) One of these things is not like the other things! Action Selection -
tmp <- starwars$birth_year - 100 starwars$`birth_year - 100` <- tmp
starwars %>% mutate(birth_year - 100) Most verbs take actions 1. New vectors are created 2. The data frame is modified -
Some verbs take selections 1. The position of columns is
looked up 2. The data frame is reorganised starwars %>% select(birth_year) tmp <- match("birth_year", colnames(starwars)) starwars[, tmp] -
starwars %>% select(c(1, height)) starwars %>% select(1:height) starwars %>% select(-1,
-height) Selections have special properties 1. c(), `-` and `:` understand positions and names 2. Selection helpers know about current variables -
starwars %>% select(ends_with("color")) starwars %>% select(matches("^[nm]a") starwars %>% select(10, everything())
1. c(), `-` and `:` understand positions and names 2. Selection helpers know about current variables Selections have special properties -
Sometimes they appear to work the same way... starwars %>%
select(height) # A tibble: 87 x 1 height <int> 1 172 2 167 3 96 # … with 84 more rows starwars %>% transmute(height) # A tibble: 87 x 1 height <int> 1 172 2 167 3 96 # … with 84 more rows -
starwars %>% select(1) # A tibble: 87 x 1 name
<chr> 1 Luke Skywalker 2 C-3PO 3 R2-D2 # … with 84 more rows starwars %>% transmute(1) # A tibble: 87 x 1 `1` <dbl> 1 1 2 1 3 1 # … with 84 more rows Sometimes they appear to work the same way... -
What about group_by()? starwars %>% group_by(gender) # A tibble: 87
x 13 # Groups: gender [5] name height mass hair_color skin_color eye_color <chr> <int> <dbl> <chr> <chr> <chr> 1 Luke… 172 77 blond fair blue 2 C-3PO 167 75 NA gold yellow 3 R2-D2 96 32 NA white, bl… red # … with 84 more rows, and 7 more variables -
starwars %>% group_by(ends_with("color")) Error: No tidyselect variables were registered What
about group_by()? It takes actions! -
What about group_by()? It takes actions! starwars %>% group_by(height >
170) %>% summarise(n()) # A tibble: 3 x 2 `height > 170` `n()` <lgl> <int> 1 FALSE 27 2 TRUE 54 3 NA 6 -
Tip: Use the _at dplyr variants to pass selections! starwars
%>% group_by_at(vars(ends_with("color"))) -
starwars %>% my_count_by(gender) # A tibble: 5 x 2 gender
n <chr> <int> 1 NA 3 2 female 19 3 hermaphrodite 1 4 male 62 5 none 2 my_count_by <- function(data, ...) { data %>% group_by(...) %>% summarise(n = n()) } Passing the dots -
starwars %>% my_count_by(GENDER = toupper(gender)) # A tibble: 5 x
2 GENDER n <chr> <int> 1 NA 3 2 FEMALE 19 3 HERMAPHRODITE 1 4 MALE 62 5 NONE 2 my_count_by <- function(data, ...) { data %>% group_by(...) %>% summarise(n = n()) } Passing the dots -
starwars %>% my_count_by(ends_with("_color")) Error: No tidyselect variables were registered my_count_by
<- function(data, ...) { data %>% group_by(...) %>% summarise(n = n()) } Passing the dots -
starwars %>% my_count_by(ends_with("_color"), -hair_color)) # A tibble: 53 x 3
# Groups: skin_color [31] skin_color eye_color n <chr> <chr> <int> 1 blue blue 1 2 blue hazel 1 3 blue, grey yellow 2 4 brown blue 1 # … with 49 more rows my_count_by <- function(data, ...) { data %>% group_by_at(vars(...)) %>% summarise(n = n()) } Passing the dots -
• Dots can be passed to aes() or vars() in
ggplot2! • They both take actions Passing the dots -
plot + facet_wrap(~ gender + hair_color) plot + facet_wrap(vars(gender, hair_color))
Facetting with formulas versus vars() • Facets historically take formulas but vars have more features • You can pass dots to vars() • vars() accepts names for facet titles -
my_wrap <- function(...) { facet_wrap(vars(...), labeller = label_both) } •
labeller controls label titles • Here, both variable name and facet category Passing the dots -
plot <- ggplot(mtcars, aes(disp, drat)) + geom_point() plot + my_wrap(
cut_number(wt, 3), cyl ) Actions! cut_number(wt, 3): (2.81,3.5] cyl: 6 cut_number(wt, 3): (2.81,3.5] cyl: 8 cut_number(wt, 3): (3.5,5.42] cyl: 8 cut_number(wt, 3): [1.51,2.81] cyl: 4 cut_number(wt, 3): [1.51,2.81] cyl: 6 cut_number(wt, 3): (2.81,3.5] cyl: 4 100 200 300 400 100 200 300 400 100 200 300 400 3.0 3.5 4.0 4.5 5.0 3.0 3.5 4.0 4.5 5.0 disp drat -
plot <- ggplot(mtcars, aes(disp, drat)) + geom_point() plot + my_wrap(
Weight = cut_number(wt, 3), Cylinder = cyl ) Weight: (2.81,3.5] Cylinder: 6 Weight: (2.81,3.5] Cylinder: 8 Weight: (3.5,5.42] Cylinder: 8 Weight: [1.51,2.81] Cylinder: 4 Weight: [1.51,2.81] Cylinder: 6 Weight: (2.81,3.5] Cylinder: 4 100 200 300 400 100 200 300 400 100 200 300 400 3.0 3.5 4.0 4.5 5.0 3.0 3.5 4.0 4.5 5.0 disp drat Named actions! -
We just wrapped a single component ⟶ Can be composed
in pipelines with + Two more examples • Entire ggplot2 pipeline • Multiple ggplot2 components Passing the dots -
Passing the dots scatter_wrap <- function(data, mapping = aes(), ...)
{ ggplot(data, mapping) + geom_point() + facet_wrap(vars(...), labeller = label_both) } Entire pipeline -
Passing the dots mtcars %>% scatter_wrap( aes(disp, drat), Cylinder =
cyl ) Cylinder: 4 Cylinder: 6 Cylinder: 8 100 200 300 400 100 200 300 400 100 200 300 400 3.0 3.5 4.0 4.5 5.0 disp drat -
Passing the dots scatter_wrap <- function(...) { geom_point() + facet_wrap(vars(...),
labeller = label_both) } Multiple components -
Passing the dots scatter_wrap <- function(...) { geom_point() + facet_wrap(vars(...),
labeller = label_both) } Multiple components An addition pipeline must start with ggplot() -
Passing the dots scatter_wrap <- function(...) { list( geom_point(), facet_wrap(vars(...),
labeller = label_both) ) } Multiple components -
Passing the dots ggplot(mtcars, aes(disp, drat)) + scatter_wrap(Cylinder = cyl)
Cylinder: 4 Cylinder: 6 Cylinder: 8 100 200 300 400 100 200 300 400 100 200 300 400 3.0 3.5 4.0 4.5 5.0 disp drat -
• Easy way of creating data masking functions • Useful
when pipeline has only one variable part • Do you need actions or selections? • See my RStudio::conf 2019 talk for more ideas Passing the dots -
1. Pass the dots 2. Subset .data 3. Interpolate Tidy
eval, the easy parts -
Subsetting .data What if you have multiple pipeline inputs? starwars
%>% group_by(gender) %>% summarise(avg = mean(mass, na.rm = TRUE)) -
.data is a pronoun that represents the data Subsetting .data
starwars %>% group_by(.data$gender) %>% summarise(avg = mean(.data$mass, na.rm = TRUE)) -
my_average <- function(data, grp_var, avg_var) { data %>% group_by(.data[[grp_var]]) %>%
summarise(avg = mean(.data[[avg_var]], na.rm = TRUE)) } Subsetting .data Just pass column names to .data[[ -
Subsetting .data starwars %>% my_average("gender", "mass") # A tibble: 5
x 2 gender avg <chr> <dbl> 1 NA 46.3 2 female 54.0 3 hermaphrodite 1358 4 male 81.0 5 none 140 Take strings! No data masking -
Subsetting .data starwars %>% my_average(gender, mass) Error: object 'gender' not
found Take strings! No data masking -
1. Pass the dots 2. Subset .data 3. Interpolate Tidy
eval, the easy parts -
• Now getting in the meat of tidy eval •
Interpolation is a simple pattern • Delay a blueprint by quoting with enquo() • Insert it back in another blueprint by unquoting with !! • Forwards a blueprint across functions Interpolation -
Simple tidy eval pattern • Delay a blueprint with enquo()
• Insert it back with !! my_average <- function(data, grp_var, avg_var) { data %>% group_by(.data[[grp_var]]) %>% summarise(avg = mean(.data[[avg_var]], na.rm = TRUE)) } Interpolation -
my_average <- function(data, grp_var, avg_var) { data %>% group_by(!!enquo(grp_var)) %>%
summarise(avg = mean(!!enquo(avg_var), na.rm = TRUE)) } Simple tidy eval pattern • Delay a blueprint with enquo() • Insert it back with !! Interpolation -
starwars %>% my_average(gender, height) # A tibble: 5 x 2
gender avg <chr> <dbl> 1 NA 120 2 female 165. 3 hermaphrodite 175 4 male 179. 5 none 200 Interpolation -
starwars %>% my_average(gender, height / 100) # A tibble: 5
x 2 gender avg <chr> <dbl> 1 NA 1.2 2 female 1.65 3 hermaphrodite 1.75 4 male 1.79 5 none 2 • Full data masking • Create vectors on the fly Interpolation -
Planned syntax: interpolation with {{ arg }} Inspiration from the
glue package thing <- "FOOBAR" glue::glue("Let's interpolate this { thing } right here") [1] "Let's interpolate this FOOBAR right here" Interpolation -
Planned syntax: interpolation with {{ arg }} my_average <- function(data,
grp_var, avg_var) { data %>% group_by(!!enquo(grp_var)) %>% summarise(avg = mean(!!enquo(avg_var), na.rm = TRUE)) } Interpolation -
Planned syntax: interpolation with {{ arg }} my_average <- function(data,
grp_var, avg_var) { data %>% group_by({{ grp_var }}) %>% summarise(avg = mean({{ avg_var }}, na.rm = TRUE)) } Interpolation -
• Simple pattern but quickly gets more complicated • What
to unquote • Delayed blueprints with enquo() • Custom blueprint material: symbols, function calls, ... • Unquoting variants such as !!! • Simple interpolation should cover many cases Interpolation -
• Data masking is a unique R feature • Great
for data analysis • Harder to program with • You might not need tidy eval • Fixed column names • Map functions on columns • Easy tidy eval techniques • Pass the dots • Subset .data • Quote and unquote (soon interpolate)