vignettes/articles/1-when-to-use-rrapply.Rmd
1-when-to-use-rrapply.Rmd
The nested list below outlines the genealogy of several famous mathematicians. Each list element contains an additional "given"
attribute with the mathematician’s given name. The numeric values at the leaf elements are the total number of descendants according to the Mathematics Genealogy Project as of June 2020. If no descendants are available there is a missing value present at the leaf node.
students <- list(
Bernoulli = structure(list(
Bernoulli = structure(list(
Bernoulli = structure(1, given = "Daniel"),
Euler = structure(list(
Euler = structure(NA, given = "Johann"),
Lagrange = structure(list(
Fourier = structure(68679, given = "Jean-Baptiste"),
Plana = structure(NA, given = "Giovanni"),
Poisson = structure(113435, given = "Simeon")
), given = "Joseph")
), given = "Leonhard")
), given = "Johann"),
Bernoulli = structure(NA, given = "Nikolaus")
), given = "Jacob")
)
str(students, give.attr = FALSE)
#> List of 1
#> $ Bernoulli:List of 2
#> ..$ Bernoulli:List of 2
#> .. ..$ Bernoulli: num 1
#> .. ..$ Euler :List of 2
#> .. .. ..$ Euler : logi NA
#> .. .. ..$ Lagrange:List of 3
#> .. .. .. ..$ Fourier: num 68679
#> .. .. .. ..$ Plana : logi NA
#> .. .. .. ..$ Poisson: num 113435
#> ..$ Bernoulli: logi NA
Consider the following exercise in list recursion:
Filter all descendants of ‘Leonhard Euler’ while keeping the original list structure and replace all missing values by zero.
Here is a possible (not so efficient) solution using recursion with the Recall()
function:
prune_replace_Euler <- function(x) {
i <- 1
while(i <= length(x)) {
if(identical(names(x)[i], "Euler") & identical(attr(x[[i]], "given"), "Leonhard")) {
x[[i]] <- rapply(x[[i]], f = function(x) replace(x, is.na(x), 0), how = "replace")
i <- i + 1
} else {
if(is.list(x[[i]])) {
val <- Recall(x[[i]])
x[[i]] <- val
i <- i + !is.null(val)
} else {
x[[i]] <- NULL
}
if(all(sapply(x, is.null))) {
x <- NULL
}
}
}
return(x)
}
str(prune_replace_Euler(students), give.attr = FALSE)
#> List of 1
#> $ Bernoulli:List of 1
#> ..$ Bernoulli:List of 1
#> .. ..$ Euler:List of 2
#> .. .. ..$ Euler : num 0
#> .. .. ..$ Lagrange:List of 3
#> .. .. .. ..$ Fourier: num 68679
#> .. .. .. ..$ Plana : num 0
#> .. .. .. ..$ Poisson: num 113435
This works, but is hardly the kind of code we would like to write for such a seemingly simple data exploration question. Moreover, this code is not very easy to follow or reason about, which makes it time-consuming to update or modify for future tasks.
Another approach would be to unnest the list in a more manageable (e.g. rectangular) format or use specialized packages such as igraph or data.tree to make pruning or modifying node entries more straightforward. Note that attention must be paid to correctly include the node attributes in the transformed object as the node names themselves are not unique in the given example. This is a sensible approach for more extensive data analysis, but ideally we would like to keep the list in its original format for simple data exploration to reduce the number of processing steps and minimize the possibility of introducing mistakes in the code.
The recursive function above makes use of rapply()
, a member of the base-R apply-family of functions, that allows to apply a function recursively to the elements of a nested list and decide how the returned result is structured. If you are not familiar with the rapply()
function it might be useful to first read the function documentation help("rapply")
or check out the first section of the rrapply
-package vignette (browseVignettes(package = "rrapply")
).
Although quite useful, the rapply()
-function is not always sufficiently flexible in practice, e.g. for pruning elements of a nested list (as demonstrated above). The rrapply()
function is an attempt to enhance and update base rapply()
to make it more generally applicable in the context of list recursion. The rrapply()
function builds on the native implementation of base rapply()
in R’s C-interface and for this reason requires no other external dependencies.
rrapply()
For illustration purposes, we will use the datasets renewable_energy_by_country
and pokedex
included in the rrapply
-package. renewable_energy_by_country
is a nested list containing the shares of renewable energy as a percentage in the total energy consumption per country in 2016. The data is publicly available at the United Nations Open SDG Data Hub UNSD-SDG07. The 249 countries and areas are structured based on their geographical location according to the United Nations M49 standard UNSD-M49 The numeric values listed for each country are percentages, if no data is available the value of the country is NA
. pokedex
is a nested list containing various property values for each of the 151 original Pokémon.
For convenience, we subset only the values for countries and areas in Oceania from renewable_energy_by_country
,
renewable_oceania <- renewable_energy_by_country[["World"]]["Oceania"]
str(renewable_oceania, list.len = 3, give.attr = FALSE)
#> List of 1
#> $ Oceania:List of 4
#> ..$ Australia and New Zealand:List of 6
#> .. ..$ Australia : num 9.32
#> .. ..$ Christmas Island : logi NA
#> .. ..$ Cocos (Keeling) Islands : logi NA
#> .. .. [list output truncated]
#> ..$ Melanesia :List of 5
#> .. ..$ Fiji : num 24.4
#> .. ..$ New Caledonia : num 4.03
#> .. ..$ Papua New Guinea: num 50.3
#> .. .. [list output truncated]
#> ..$ Micronesia :List of 8
#> .. ..$ Guam : num 3.03
#> .. ..$ Kiribati : num 45.4
#> .. ..$ Marshall Islands : num 11.8
#> .. .. [list output truncated]
#> .. [list output truncated]
With base rapply()
, there is no convenient way to prune or filter elements from the input list. The rrapply()
function adds an option how = "prune"
to prune all list elements not subject to application of the function f
from a nested list. The original list structure is retained, similar to the non-pruned versions how = "replace"
and how = "list"
. Using how = "prune"
, we can for instance drop all NA
elements from the list while preserving the original list structure:
## Drop all logical NA's while preserving list structure
na_drop_oceania <- rrapply(
renewable_oceania,
f = identity,
classes = "numeric",
how = "prune"
)
str(na_drop_oceania, list.len = 3, give.attr = FALSE)
#> List of 1
#> $ Oceania:List of 4
#> ..$ Australia and New Zealand:List of 2
#> .. ..$ Australia : num 9.32
#> .. ..$ New Zealand: num 32.8
#> ..$ Melanesia :List of 5
#> .. ..$ Fiji : num 24.4
#> .. ..$ New Caledonia : num 4.03
#> .. ..$ Papua New Guinea: num 50.3
#> .. .. [list output truncated]
#> ..$ Micronesia :List of 7
#> .. ..$ Guam : num 3.03
#> .. ..$ Kiribati : num 45.4
#> .. ..$ Marshall Islands : num 11.8
#> .. .. [list output truncated]
#> .. [list output truncated]
Instead, set how = "flatten"
to return a flattened unnested version of the pruned list. This is more efficient than first returning the pruned list with how = "prune"
and unlisting or flattening the list in a subsequent step.
## Drop all logical NA's and return unnested list
na_drop_oceania2 <- rrapply(
renewable_oceania,
f = identity,
classes = "numeric",
how = "flatten"
)
head(na_drop_oceania2, n = 10)
#> Australia New Zealand Fiji New Caledonia
#> 9.32 32.76 24.36 4.03
#> Papua New Guinea Solomon Islands Vanuatu Guam
#> 50.34 65.73 33.67 3.03
#> Kiribati Marshall Islands
#> 45.43 11.75
Or, use how = "melt"
to return a melted data.frame of the pruned list similar in format to reshape2::melt()
applied to a nested list. The rows of the melted data.frame contain the node paths of the elements in the pruned list. The "value"
column contains the values of the terminal nodes analogous to the flattened list returned by how = "flatten"
.
## Drop all logical NA's and return melted data.frame
na_drop_oceania3 <- rrapply(
renewable_oceania,
f = identity,
classes = "numeric",
how = "melt"
)
head(na_drop_oceania3)
#> L1 L2 L3 value
#> 1 Oceania Australia and New Zealand Australia 9.32
#> 2 Oceania Australia and New Zealand New Zealand 32.76
#> 3 Oceania Melanesia Fiji 24.36
#> 4 Oceania Melanesia New Caledonia 4.03
#> 5 Oceania Melanesia Papua New Guinea 50.34
#> 6 Oceania Melanesia Solomon Islands 65.73
If no names are present in a sublist of the input list, how = "melt"
replaces the names in the melted data.frame by list element indices "1"
, "2"
, etc.:
## Remove all names at L2
## (skip this for now, these arguments are explained in the following sections)
oceania_unnamed <- rrapply(
renewable_oceania,
classes = "list",
condition = function(x, .xname) .xname == "Oceania",
f = unname
)
## Drop all logical NA's and return melted data.frame
na_drop_oceania4 <- rrapply(
oceania_unnamed,
f = identity,
classes = "numeric",
how = "melt"
)
head(na_drop_oceania4)
#> L1 L2 L3 value
#> 1 Oceania 1 Australia 9.32
#> 2 Oceania 1 New Zealand 32.76
#> 3 Oceania 2 Fiji 24.36
#> 4 Oceania 2 New Caledonia 4.03
#> 5 Oceania 2 Papua New Guinea 50.34
#> 6 Oceania 2 Solomon Islands 65.73
Nested lists containing repeated observations, such as pokedex
, can be unnested with how = "bind"
, which expands and binds the repeated observations (i.e. sublists) as individual rows in a wide data.frame. This is similar in format to repeated application of tidyr::unnest_wider()
to a nested data.frame.
## Nested list of Pokemon properties in Pokemon GO
data("pokedex")
str(pokedex, list.len = 3)
#> List of 1
#> $ pokemon:List of 151
#> ..$ :List of 16
#> .. ..$ id : int 1
#> .. ..$ num : chr "001"
#> .. ..$ name : chr "Bulbasaur"
#> .. .. [list output truncated]
#> ..$ :List of 17
#> .. ..$ id : int 2
#> .. ..$ num : chr "002"
#> .. ..$ name : chr "Ivysaur"
#> .. .. [list output truncated]
#> ..$ :List of 15
#> .. ..$ id : int 3
#> .. ..$ num : chr "003"
#> .. ..$ name : chr "Venusaur"
#> .. .. [list output truncated]
#> .. [list output truncated]
## Unnest list as a wide data.frame
pokedex_wide <- rrapply(pokedex, how = "bind")
head(pokedex_wide[, c(1:3, 5:10)], n = 5)
#> pokemon.id pokemon.num pokemon.name pokemon.type pokemon.height
#> 1 1 001 Bulbasaur Grass, Poison 0.71 m
#> 2 2 002 Ivysaur Grass, Poison 0.99 m
#> 3 3 003 Venusaur Grass, Poison 2.01 m
#> 4 4 004 Charmander Fire 0.61 m
#> 5 5 005 Charmeleon Fire 1.09 m
#> pokemon.weight pokemon.candy pokemon.candy_count pokemon.egg
#> 1 6.9 kg Bulbasaur Candy 25 2 km
#> 2 13.0 kg Bulbasaur Candy 100 Not in Eggs
#> 3 100.0 kg Bulbasaur Candy NA Not in Eggs
#> 4 8.5 kg Charmander Candy 25 2 km
#> 5 19.0 kg Charmander Candy 100 Not in Eggs
Base rapply()
allows to apply a function f
to list elements of certain types or classes via the classes
argument. rrapply()
generalizes this concept via the condition
argument, which accepts any principal argument function to use as a condition or predicate to apply f
to a subset of list elements. Conceptually, the f
function is applied to all leaf elements for which the condition
function exactly evaluates to TRUE
similar to isTRUE()
. If the condition
argument is missing, f
is applied to all leaf elements. In combination with how = "prune"
, the condition
function provides additional flexibility in selecting and filtering elements from a nested list. Using the condition
argument, we can update the above function call to drop all NA
’s from the list to better reflect our purpose:
## drop all NA elements using condition function
na_drop_oceania3 <- rrapply(
renewable_oceania,
condition = Negate(is.na),
f = identity,
how = "prune"
)
str(na_drop_oceania3, list.len = 3, give.attr = FALSE)
#> List of 1
#> $ Oceania:List of 4
#> ..$ Australia and New Zealand:List of 2
#> .. ..$ Australia : num 9.32
#> .. ..$ New Zealand: num 32.8
#> ..$ Melanesia :List of 5
#> .. ..$ Fiji : num 24.4
#> .. ..$ New Caledonia : num 4.03
#> .. ..$ Papua New Guinea: num 50.3
#> .. .. [list output truncated]
#> ..$ Micronesia :List of 7
#> .. ..$ Guam : num 3.03
#> .. ..$ Kiribati : num 45.4
#> .. ..$ Marshall Islands : num 11.8
#> .. .. [list output truncated]
#> .. [list output truncated]
More interesting is to consider a condition
that cannot also be defined using the classes
argument. For instance, we can filter all countries with values that satisfy a certain numeric condition:
## filter all countries with values above 85%
renewable_energy_above_85 <- rrapply(
renewable_energy_by_country,
condition = function(x) x > 85,
how = "prune"
)
str(renewable_energy_above_85, give.attr = FALSE)
#> List of 1
#> $ World:List of 1
#> ..$ Africa:List of 1
#> .. ..$ Sub-Saharan Africa:List of 3
#> .. .. ..$ Eastern Africa:List of 7
#> .. .. .. ..$ Burundi : num 89.2
#> .. .. .. ..$ Ethiopia : num 91.9
#> .. .. .. ..$ Rwanda : num 86
#> .. .. .. ..$ Somalia : num 94.7
#> .. .. .. ..$ Uganda : num 88.6
#> .. .. .. ..$ United Republic of Tanzania: num 86.1
#> .. .. .. ..$ Zambia : num 88.5
#> .. .. ..$ Middle Africa :List of 2
#> .. .. .. ..$ Chad : num 85.3
#> .. .. .. ..$ Democratic Republic of the Congo: num 97
#> .. .. ..$ Western Africa:List of 1
#> .. .. .. ..$ Guinea-Bissau: num 86.5
## or by passing arguments to condition via ...
renewable_energy_equal_0 <- rrapply(
renewable_energy_by_country,
condition = "==",
e2 = 0,
how = "prune"
)
str(renewable_energy_equal_0, give.attr = FALSE)
#> List of 1
#> $ World:List of 4
#> ..$ Americas:List of 1
#> .. ..$ Latin America and the Caribbean:List of 1
#> .. .. ..$ Caribbean:List of 1
#> .. .. .. ..$ Antigua and Barbuda: num 0
#> ..$ Asia :List of 1
#> .. ..$ Western Asia:List of 4
#> .. .. ..$ Bahrain: num 0
#> .. .. ..$ Kuwait : num 0
#> .. .. ..$ Oman : num 0
#> .. .. ..$ Qatar : num 0
#> ..$ Europe :List of 2
#> .. ..$ Northern Europe:List of 1
#> .. .. ..$ Channel Islands:List of 1
#> .. .. .. ..$ Guernsey: num 0
#> .. ..$ Southern Europe:List of 1
#> .. .. ..$ Gibraltar: num 0
#> ..$ Oceania :List of 2
#> .. ..$ Micronesia:List of 1
#> .. .. ..$ Northern Mariana Islands: num 0
#> .. ..$ Polynesia :List of 1
#> .. .. ..$ Wallis and Futuna Islands: num 0
Remark: Note that the NA
elements are not returned, as the condition
function does not evaluate to TRUE
for NA
values. Also, the f
argument is allowed to be missing, in which case no function is applied to the leaf elements.
As the condition
function is a generalization of the classes
argument to have more flexible control of the predicate, it is also possible to use the deflt
argument together with how = "list"
or how = "unlist"
to set a default value to all leaf elements for which the condition
does not evaluate to TRUE
:
## replace all NA elements by zero
na_zero_oceania_list <- rrapply(
renewable_oceania,
condition = Negate(is.na),
deflt = 0,
how = "list"
)
str(na_zero_oceania_list, list.len = 3, give.attr = FALSE)
#> List of 1
#> $ Oceania:List of 4
#> ..$ Australia and New Zealand:List of 6
#> .. ..$ Australia : num 9.32
#> .. ..$ Christmas Island : num 0
#> .. ..$ Cocos (Keeling) Islands : num 0
#> .. .. [list output truncated]
#> ..$ Melanesia :List of 5
#> .. ..$ Fiji : num 24.4
#> .. ..$ New Caledonia : num 4.03
#> .. ..$ Papua New Guinea: num 50.3
#> .. .. [list output truncated]
#> ..$ Micronesia :List of 8
#> .. ..$ Guam : num 3.03
#> .. ..$ Kiribati : num 45.4
#> .. ..$ Marshall Islands : num 11.8
#> .. .. [list output truncated]
#> .. [list output truncated]
To be consistent with base rapply()
, the deflt
argument can still only be used together with how = "list"
or how = "unlist"
.
...
argumentThe first argument to f
always evaluates to the content of the leaf element to which f
is applied. Any further arguments (besides the special arguments .xname
, .xpos
, .xparents
and .xsiblings
discussed below) that are independent of the node content can be supplied via the ...
argument. Since rrapply()
accepts a function in two of its arguments f
and condition
, any further arguments defined via the ...
also need to be defined as function arguments in both the f
and condition
function (if existing), even if they are not used in the function itself.
To clarify, consider the following example where we replace all NA
elements by a value defined in a separate argument newvalue
:
## this is not ok!
tryCatch({
rrapply(
renewable_oceania,
condition = is.na,
f = function(x, newvalue) newvalue,
newvalue = 0,
how = "replace"
)
}, error = function(error) error$message)
#> [1] "2 arguments passed to 'is.na' which requires 1"
## this is ok
na_zero_oceania_replace3 <- rrapply(
renewable_oceania,
condition = function(x, newvalue) is.na(x),
f = function(x, newvalue) newvalue,
newvalue = 0,
how = "replace"
)
str(na_zero_oceania_replace3, list.len = 3, give.attr = FALSE)
#> List of 1
#> $ Oceania:List of 4
#> ..$ Australia and New Zealand:List of 6
#> .. ..$ Australia : num 9.32
#> .. ..$ Christmas Island : num 0
#> .. ..$ Cocos (Keeling) Islands : num 0
#> .. .. [list output truncated]
#> ..$ Melanesia :List of 5
#> .. ..$ Fiji : num 24.4
#> .. ..$ New Caledonia : num 4.03
#> .. ..$ Papua New Guinea: num 50.3
#> .. .. [list output truncated]
#> ..$ Micronesia :List of 8
#> .. ..$ Guam : num 3.03
#> .. ..$ Kiribati : num 45.4
#> .. ..$ Marshall Islands : num 11.8
#> .. .. [list output truncated]
#> .. [list output truncated]
.xname
, .xpos
, .xparents
and .xsiblings
In base rapply()
, the f
function only has access to the content of the list element under evaluation through its principal argument, and there is no convenient way to access its name or location in the nested list from inside the f
function. This makes rapply()
impractical if we want to apply a function f
that relies on e.g. the name or position of a node as in the students
example above. To overcome this limitation, rrapply()
allows the use of the special arguments .xname
, .xpos
, .xparents
and .xsiblings
inside the f
and condition
functions (in addition to the principal function argument). .xname
evaluates to the name of the list element, .xpos
evaluates to the position of the element in the nested list structured as an integer vector, .xparents
evaluates to a vector of all parent node names in the path to the current list element, and .xsiblings
evaluates to the parent list containing the current list element and all of its direct siblings.
Using the .xname
and .xpos
arguments, we can transform or filter list elements based on their names or positions in the nested list:
## apply a function using the name of the node
renewable_oceania_text <- rrapply(
renewable_oceania,
f = function(x, .xname) sprintf("Renewable energy in %s: %.2f%%", .xname, x),
condition = Negate(is.na),
how = "flatten"
)
head(renewable_oceania_text, n = 5)
#> Australia
#> "Renewable energy in Australia: 9.32%"
#> New Zealand
#> "Renewable energy in New Zealand: 32.76%"
#> Fiji
#> "Renewable energy in Fiji: 24.36%"
#> New Caledonia
#> "Renewable energy in New Caledonia: 4.03%"
#> Papua New Guinea
#> "Renewable energy in Papua New Guinea: 50.34%"
## extract values based on country names
renewable_benelux <- rrapply(
renewable_energy_by_country,
condition = function(x, .xname) .xname %in% c("Belgium", "Netherlands", "Luxembourg"),
how = "prune"
)
str(renewable_benelux, give.attr = FALSE)
#> List of 1
#> $ World:List of 1
#> ..$ Europe:List of 1
#> .. ..$ Western Europe:List of 3
#> .. .. ..$ Belgium : num 9.14
#> .. .. ..$ Luxembourg : num 13.5
#> .. .. ..$ Netherlands: num 5.78
Knowing that Europe is located under the node renewable_energy_by_country[[c(1, 5)]]
, we can filter all European countries with a renewable energy share above 50 percent by using the .xpos
argument,
## filter European countries with value above 50%
renewable_europe_above_50 <- rrapply(
renewable_energy_by_country,
condition = function(x, .xpos) identical(head(.xpos, 2), c(1L, 5L)) & x > 50,
how = "prune"
)
str(renewable_europe_above_50, give.attr = FALSE)
#> List of 1
#> $ World:List of 1
#> ..$ Europe:List of 2
#> .. ..$ Northern Europe:List of 3
#> .. .. ..$ Iceland: num 78.1
#> .. .. ..$ Norway : num 59.5
#> .. .. ..$ Sweden : num 51.4
#> .. ..$ Western Europe :List of 1
#> .. .. ..$ Liechtenstein: num 62.9
This can be done more conveniently using the .xparents
argument, as this does not require us to look up the location of Europe in the list beforehand,
## filter European countries with value above 50%
renewable_europe_above_50 <- rrapply(
renewable_energy_by_country,
condition = function(x, .xparents) "Europe" %in% .xparents & x > 50,
how = "prune"
)
str(renewable_europe_above_50, give.attr = FALSE)
#> List of 1
#> $ World:List of 1
#> ..$ Europe:List of 2
#> .. ..$ Northern Europe:List of 3
#> .. .. ..$ Iceland: num 78.1
#> .. .. ..$ Norway : num 59.5
#> .. .. ..$ Sweden : num 51.4
#> .. ..$ Western Europe :List of 1
#> .. .. ..$ Liechtenstein: num 62.9
Using the .xpos
argument, we could look up the location of a particular country in the nested list,
## look up position of Sweden in list
(xpos_sweden <- rrapply(
renewable_energy_by_country,
condition = function(x, .xname) identical(.xname, "Sweden"),
f = function(x, .xpos) .xpos,
how = "flatten"
))
#> $Sweden
#> [1] 1 5 2 14
renewable_energy_by_country[[xpos_sweden$Sweden]]
#> [1] 51.35
#> attr(,"M49-code")
#> [1] "752"
Instead, using the .xsiblings
argument we could look up the direct neighbors of a particular country in the nested list,
## look up sibling countries of Sweden in list
siblings_sweden <- rrapply(
renewable_energy_by_country,
condition = function(x, .xsiblings) "Sweden" %in% names(.xsiblings),
how = "flatten"
)
head(siblings_sweden, n = 10)
#> Aland Islands Denmark Estonia Faroe Islands Finland
#> NA 33.06 26.55 4.24 42.03
#> Iceland Ireland Isle of Man Latvia Lithuania
#> 78.07 8.65 4.30 38.48 31.42
We could also use the .xpos
argument to determine the maximum depth of the list or the length of the longest sublist as follows,
## maximum list depth
depth_all <- rrapply(
renewable_energy_by_country,
f = function(x, .xpos) length(.xpos),
how = "unlist"
)
max(depth_all)
#> [1] 5
## longest sublist length
sublist_count <- rrapply(
renewable_energy_by_country,
f = function(x, .xpos) max(.xpos),
how = "unlist"
)
max(sublist_count)
#> [1] 28
When unnesting nested lists with how = "bind"
, the .xname
, .xpos
or .xparents
arguments can be useful to parse only a particular set of data.frame columns:
## parse only Pokemon number, name and type columns
pokedex_small <- rrapply(
pokedex,
condition = function(x, .xpos, .xname) {
length(.xpos) < 4 & .xname %in% c("num", "name", "type")
},
how = "bind"
)
head(pokedex_small)
#> pokemon.num pokemon.name pokemon.type
#> 1 001 Bulbasaur Grass, Poison
#> 2 002 Ivysaur Grass, Poison
#> 3 003 Venusaur Grass, Poison
#> 4 004 Charmander Fire
#> 5 005 Charmeleon Fire
#> 6 006 Charizard Fire, Flying
By default, if classes = "ANY"
both base rapply()
and rrapply()
recurse into any list-like element. Using classes = "list"
in base rapply()
has no effect as the function descends into any list node before evaluating the classes
argument. In contrast, rrapply()
does detect classes = "list"
, in which case the f
function is applied to list elements that satisfy the condition
function. If the condition
is not satisfied for a list element, rrapply()
will recurse further into the sublist, apply the f
function to the nodes that satisfy condition
and so on. The use of classes = "list"
tells rrapply()
to not descend into list objects by default. For this reason this behavior can only be triggered with the classes
argument and not through the use of e.g. condition = is.list
.
The choice classes = "list"
is useful to calculate summary statistics across nodes or to look up the position of intermediate nodes in a nested list. To illustrate, we can return the mean and standard deviation of the renewable energy share in Europe as follows:
## compute mean value of Europe
rrapply(
renewable_energy_by_country,
classes = "list",
condition = function(x, .xname) .xname == "Europe",
f = function(x) list(
mean = mean(unlist(x), na.rm = TRUE),
sd = sd(unlist(x), na.rm = TRUE)
),
how = "flatten"
)
#> $Europe
#> $Europe$mean
#> [1] 22.36565
#>
#> $Europe$sd
#> [1] 17.12639
Remark: Note that the principal x
argument in the f
function now evaluates to a list for the node satisfying the condition
. For this reason, we first unlist
the sublist before passing it to mean
and sd
.
We can use the .xpos
argument to apply the f
function only at specific locations or depths in the nested list. For instance, we could return the mean renewable energy shares for each continent by making use of the fact that the .xpos
vector of each continent has length (i.e. depth) 2:
## compute mean value of each continent
renewable_continent_summary <- rrapply(
renewable_energy_by_country,
classes = "list",
condition = function(x, .xpos) length(.xpos) == 2,
f = function(x) mean(unlist(x), na.rm = TRUE)
)
## Antarctica has a missing value
str(renewable_continent_summary, give.attr = FALSE)
#> List of 1
#> $ World:List of 6
#> ..$ Africa : num 54.3
#> ..$ Americas : num 18.2
#> ..$ Antarctica: logi NA
#> ..$ Asia : num 17.9
#> ..$ Europe : num 22.4
#> ..$ Oceania : num 17.8
If classes = "list"
, the f
function is only applied to the (non-terminal) list nodes. If classes = "ANY"
, the f
function is only applied to any (terminal) non-list node. To apply f
to any terminal and non-terminal node of the nested list, we should combine classes = c("list", "ANY")
. This is illustrated by searching both terminal and non-terminal nodes for the country or region with M49-code "155"
(Western Europe):
## Filter country or region by M49-code
rrapply(
renewable_energy_by_country,
classes = c("list", "ANY"),
condition = function(x) attr(x, "M49-code") == "155",
f = function(x, .xname) .xname,
how = "unlist"
)
#> World.Europe.Western Europe
#> "Western Europe"
Another illustrating example is to preprocess certain list nodes before unnesting the nested list. Here, we extract the Pokémon evolutions in pokedex
with and without pre-processing of the Pokémon evolution list nodes:
## Pokemon evolution columns without pre-processing
pokedex_wide1 <- rrapply(
pokedex,
condition = function(x, .xparents) any(grepl("name|evolution", .xparents)),
how = "bind"
)
head(pokedex_wide1, n = 3)
#> pokemon.name pokemon.next_evolution.1.num pokemon.next_evolution.1.name
#> 1 Bulbasaur 002 Ivysaur
#> 2 Ivysaur 003 Venusaur
#> 3 Venusaur <NA> <NA>
#> pokemon.next_evolution.2.num pokemon.next_evolution.2.name
#> 1 003 Venusaur
#> 2 <NA> <NA>
#> 3 <NA> <NA>
#> pokemon.prev_evolution.1.num pokemon.prev_evolution.1.name
#> 1 <NA> <NA>
#> 2 001 Bulbasaur
#> 3 001 Bulbasaur
#> pokemon.prev_evolution.2.num pokemon.prev_evolution.2.name
#> 1 <NA> <NA>
#> 2 <NA> <NA>
#> 3 002 Ivysaur
#> pokemon.next_evolution.3.num pokemon.next_evolution.3.name
#> 1 <NA> <NA>
#> 2 <NA> <NA>
#> 3 <NA> <NA>
## Pokemon evolution columns simplified to character vectors
pokedex_wide2 <- rrapply(
pokedex,
classes = c("character", "list"),
condition = function(x, .xparents) any(grepl("name|evolution", .xparents)),
f = function(x) if(is.list(x)) sapply(x, `[[`, "name") else x,
how = "bind"
)
head(pokedex_wide2, n = 9)
#> pokemon.name pokemon.next_evolution pokemon.prev_evolution
#> 1 Bulbasaur Ivysaur, Venusaur NA
#> 2 Ivysaur Venusaur Bulbasaur
#> 3 Venusaur NA Bulbasaur, Ivysaur
#> 4 Charmander Charmeleon, Charizard NA
#> 5 Charmeleon Charizard Charmander
#> 6 Charizard NA Charmander, Charmeleon
#> 7 Squirtle Wartortle, Blastoise NA
#> 8 Wartortle Blastoise Squirtle
#> 9 Blastoise NA Squirtle, Wartortle
If classes = "list"
, rrapply()
applies the f
function to any list element that satisfies the condition
function, but will not recurse further into these list elements. This makes it for instance impossible to recursively update the name of each list element in a nested list, as rrapply()
stops recursing after updating the first list layer. For this purpose, set classes = "list"
combined with how = "recurse"
, in which case rrapply()
recurses further into any updated list element after application of the f
function (using how = "replace"
). In this context, the condition
argument is interpreted as a passing criterion: as long as the condition
and classes
arguments are satisfied, rrapply()
will try to recurse further into any list-like element.
Using how = "recurse"
, we can recursively replace all names in renewable_energy_by_country
by their M49-codes:
## replace country names by M-49 codes
renewable_M49 <- rrapply(
list(renewable_energy_by_country),
classes = "list",
f = function(x) {
names(x) <- vapply(x, attr, character(1L), which = "M49-code")
return(x)
},
how = "recurse"
)
str(renewable_M49[[1]], max.level = 3, list.len = 3, give.attr = FALSE)
#> List of 1
#> $ 001:List of 6
#> ..$ 002:List of 2
#> .. ..$ 015:List of 7
#> .. ..$ 202:List of 4
#> ..$ 019:List of 2
#> .. ..$ 419:List of 3
#> .. ..$ 021:List of 5
#> ..$ 010: logi NA
#> .. [list output truncated]
Remark: Here we passed list(renewable_energy_by_country)
to the call of rrapply()
in order to start application of the f
function at the level of the list renewable_energy_by_country
itself, instead of starting at its list elements.
If classes = "ANY"
, rrapply()
recurses into any list-like object equivalent to base rapply()
. Since data.frames are list-like objects, the f
function is applied to the individual columns instead of the data.frame object as a whole. To avoid this behavior, set classes = "data.frame"
, in which case the f
and condition
functions are applied directly to the data.frame object itself and not its columns. Note that with base rapply()
using classes = "data.frame"
has no effect as rapply()
descends into the columns of any data.frame object before evaluating the classes
argument.
## create a list of data.frames
oceania_df <- list(
Oceania = lapply(
renewable_oceania[["Oceania"]],
FUN = function(x) data.frame(
Name = names(x),
value = unlist(x),
stringsAsFactors = FALSE
)
)
)
## this does not work!
tryCatch({
rrapply(
oceania_df,
f = function(x) subset(x, !is.na(value)), ## filter NA-rows of data.frame
how = "replace"
)
}, error = function(error) error$message)
#> [1] "object 'value' not found"
## this does work
rrapply(
oceania_df,
classes = "data.frame",
f = function(x) subset(x, !is.na(value)),
how = "replace"
)
#> $Oceania
#> $Oceania$`Australia and New Zealand`
#> Name value
#> Australia Australia 9.32
#> New Zealand New Zealand 32.76
#>
#> $Oceania$Melanesia
#> Name value
#> Fiji Fiji 24.36
#> New Caledonia New Caledonia 4.03
#> Papua New Guinea Papua New Guinea 50.34
#> Solomon Islands Solomon Islands 65.73
#> Vanuatu Vanuatu 33.67
#>
#> $Oceania$Micronesia
#> Name value
#> Guam Guam 3.03
#> Kiribati Kiribati 45.43
#> Marshall Islands Marshall Islands 11.75
#> Micronesia (Federated States of) Micronesia (Federated States of) 1.64
#> Nauru Nauru 31.44
#> Northern Mariana Islands Northern Mariana Islands 0.00
#> Palau Palau 0.02
#>
#> $Oceania$Polynesia
#> Name value
#> American Samoa American Samoa 1.00
#> Cook Islands Cook Islands 1.90
#> French Polynesia French Polynesia 11.06
#> Niue Niue 22.07
#> Samoa Samoa 27.30
#> Tonga Tonga 1.98
#> Tuvalu Tuvalu 11.76
#> Wallis and Futuna Islands Wallis and Futuna Islands 0.00
Remark: Note that the same result can also be obtained using classes = "list"
and checking that the list element under evaluation is a data.frame:
rrapply(
oceania_df,
classes = "list",
condition = is.data.frame,
f = function(x) subset(x, !is.na(value)),
how = "replace"
)
Base rapply()
may produce different results when using how = "replace"
or how = "list"
when working with list attributes. The former preserves intermediate list attributes whereas the latter does not. To avoid unexpected behavior, rrapply()
always preserves intermediate list attributes when using how = "replace"
, how = "list"
or how = "prune"
. If how = "flatten"
, how = "melt"
, how = "bind"
or how = "unlist"
intermediate list attributes cannot be preserved as the result is no longer a nested list.
## how = "list" now preserves all list attributes
na_drop_oceania_list_attr2 <- rrapply(
renewable_oceania,
f = function(x) replace(x, is.na(x), 0),
how = "list"
)
str(na_drop_oceania_list_attr2, max.level = 2)
#> List of 1
#> $ Oceania:List of 4
#> ..$ Australia and New Zealand:List of 6
#> .. ..- attr(*, "M49-code")= chr "053"
#> ..$ Melanesia :List of 5
#> .. ..- attr(*, "M49-code")= chr "054"
#> ..$ Micronesia :List of 8
#> .. ..- attr(*, "M49-code")= chr "057"
#> ..$ Polynesia :List of 10
#> .. ..- attr(*, "M49-code")= chr "061"
#> ..- attr(*, "M49-code")= chr "009"
## how = "prune" also preserves list attributes
na_drop_oceania_attr <- rrapply(
renewable_oceania,
condition = Negate(is.na),
how = "prune"
)
str(na_drop_oceania_attr, max.level = 2)
#> List of 1
#> $ Oceania:List of 4
#> ..$ Australia and New Zealand:List of 2
#> .. ..- attr(*, "M49-code")= chr "053"
#> ..$ Melanesia :List of 5
#> .. ..- attr(*, "M49-code")= chr "054"
#> ..$ Micronesia :List of 7
#> .. ..- attr(*, "M49-code")= chr "057"
#> ..$ Polynesia :List of 8
#> .. ..- attr(*, "M49-code")= chr "061"
#> ..- attr(*, "M49-code")= chr "009"
rrapply()
on data.framesThe previous sections explained how to avoid recursing into list-like elements using rrapply()
. However, it can also be useful to exploit the property that a data.frame is a list-like object and use base rapply()
to apply a function f
to data.frame columns of certain classes. For instance, it is straightforward to standardize all numeric
columns in the iris
dataset by their sample mean and standard deviation:
iris_standard <- rapply(iris, f = scale, classes = "numeric", how = "replace")
head(iris_standard)
#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> 1 -0.8976739 1.01560199 -1.335752 -1.311052 setosa
#> 2 -1.1392005 -0.13153881 -1.335752 -1.311052 setosa
#> 3 -1.3807271 0.32731751 -1.392399 -1.311052 setosa
#> 4 -1.5014904 0.09788935 -1.279104 -1.311052 setosa
#> 5 -1.0184372 1.24503015 -1.335752 -1.311052 setosa
#> 6 -0.5353840 1.93331463 -1.165809 -1.048667 setosa
Using the condition
argument in rrapply()
, we gain additional flexibility in selecting the columns to which f
is applied. For instance, we can apply the f
function only to the Sepal
columns using the .xname
argument:
iris_standard_sepal <- rrapply(
iris,
condition = function(x, .xname) grepl("Sepal", .xname),
f = scale
)
head(iris_standard_sepal)
#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> 1 -0.8976739 1.01560199 1.4 0.2 setosa
#> 2 -1.1392005 -0.13153881 1.4 0.2 setosa
#> 3 -1.3807271 0.32731751 1.3 0.2 setosa
#> 4 -1.5014904 0.09788935 1.5 0.2 setosa
#> 5 -1.0184372 1.24503015 1.4 0.2 setosa
#> 6 -0.5353840 1.93331463 1.7 0.4 setosa
Instead of mutating columns, we can also transmute columns (i.e. keeping only the columns to which f
is applied) by setting how = "prune"
:
iris_standard_transmute <- rrapply(
iris,
f = scale,
classes = "numeric",
how = "prune"
)
head(iris_standard_transmute)
#> Sepal.Length Sepal.Width Petal.Length Petal.Width
#> 1 -0.8976739 1.01560199 -1.335752 -1.311052
#> 2 -1.1392005 -0.13153881 -1.335752 -1.311052
#> 3 -1.3807271 0.32731751 -1.392399 -1.311052
#> 4 -1.5014904 0.09788935 -1.279104 -1.311052
#> 5 -1.0184372 1.24503015 -1.335752 -1.311052
#> 6 -0.5353840 1.93331463 -1.165809 -1.048667
In order to summarize a set of selected columns, use how = "flatten"
instead of how = "prune"
, as the latter preserves list attributes –including data.frame dimensions– which should not be kept.
## summarize columns with how = "flatten"
iris_standard_summarize <- rrapply(
iris,
f = summary,
classes = "numeric",
how = "flatten"
)
iris_standard_summarize
#> $Sepal.Length
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> 4.300 5.100 5.800 5.843 6.400 7.900
#>
#> $Sepal.Width
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> 2.000 2.800 3.000 3.057 3.300 4.400
#>
#> $Petal.Length
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> 1.000 1.600 4.350 3.758 5.100 6.900
#>
#> $Petal.Width
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> 0.100 0.300 1.300 1.199 1.800 2.500
To conclude, let us return to the list recursion exercise in the first section. Using rrapply()
, we can solve the task in a more readable and less error-prone fashion. A possible approach is to split up the question in two steps as follows:
## look up the position of Euler (Leonhard)
xpos_Euler <- rrapply(
students,
classes = "list",
condition = function(x, .xname) .xname == "Euler" && attr(x, "given") == "Leonhard",
f = function(x, .xpos) .xpos,
how = "flatten"
)[[1]]
xpos_Euler
#> [1] 1 1 2
## filter descendants of Euler (Leonhard) and replace missing values by zero
students_Euler <- rrapply(
students,
condition = function(x, .xpos) identical(.xpos[seq_along(xpos_Euler)], xpos_Euler),
f = function(x) replace(x, is.na(x), 0),
how = "prune"
)
str(students_Euler, give.attr = FALSE)
#> List of 1
#> $ Bernoulli:List of 1
#> ..$ Bernoulli:List of 1
#> .. ..$ Euler:List of 2
#> .. .. ..$ Euler : num 0
#> .. .. ..$ Lagrange:List of 3
#> .. .. .. ..$ Fourier: num 68679
#> .. .. .. ..$ Plana : num 0
#> .. .. .. ..$ Poisson: num 113435
Knowing that Johann Euler is a descendant of Leonhard Euler, we can further simplify this into a single function call using the .xparents
argument:
## filter descendants of Euler (Leonhard) and replace missing values by zero
students_Euler <- rrapply(
students,
condition = function(x, .xparents) "Euler" %in% .xparents,
f = function(x) replace(x, is.na(x), 0),
how = "prune"
)
str(students_Euler, give.attr = FALSE)
#> List of 1
#> $ Bernoulli:List of 1
#> ..$ Bernoulli:List of 1
#> .. ..$ Euler:List of 2
#> .. .. ..$ Euler : num 0
#> .. .. ..$ Lagrange:List of 3
#> .. .. .. ..$ Fourier: num 68679
#> .. .. .. ..$ Plana : num 0
#> .. .. .. ..$ Poisson: num 113435
For additional information and details check out the package vignette (browseVignettes(package = "rrapply")
) or the other articles in the Articles section.