Handling NULL after if() inside appply - r

I have a situation where apply returns a list with many NULL entries. The code I use is quite long, so I reproduce the problem with a simple example.
# Generate data.
df <- data.frame(a= c(1,2, NA, 6),
b= c(1, 7, 3, 7))
# Return only columns that have NAs.
my_list <- apply(df, 2, function(col_i){
if(any(is.na(col_i))){
return(col_i)
}})
Running this gives us
my_list
$a
[1] 1 2 NA 6
$b
NULL
My problem is that I get many Null entries so that I can not work with the results. How can I (a) avoid apply to return the NULL entries or (b) discard all NULL entries in my_list?
So the expected output is
my_list
$a
[1] 1 2 NA 6
Again, the actual code I use is more complex than that. So please do not suggest using something like df[ , !complete.cases(t(df)), drop= FALSE] which also returns columns that contain any missings. My question is not about how to get columns with any missings but how to handle NULLentries in apply. I want to keep the if part inside apply.


Every function has to return something so you can't really avoid returning something but you can remove them :
Filter(length, apply(df, 2, function(col_i) if(any(is.na(col_i))) return(col_i)))
#$a
#[1] 1 2 NA 6

Related

Why ifelse or if_else doesn't work as expected in purrr::accumulate [duplicate]

I've found R's ifelse statements to be pretty handy from time to time. For example:
ifelse(TRUE,1,2)
# [1] 1
ifelse(FALSE,1,2)
# [1] 2
But I'm somewhat confused by the following behavior.
ifelse(TRUE,c(1,2),c(3,4))
# [1] 1
ifelse(FALSE,c(1,2),c(3,4))
# [1] 3
Is this a design choice that's above my paygrade?

The documentation for ifelse states:
ifelse returns a value with the same
shape as test which is filled with
elements selected from either yes or
no depending on whether the element
of test is TRUE or FALSE.
Since you are passing test values of length 1, you are getting results of length 1. If you pass longer test vectors, you will get longer results:
> ifelse(c(TRUE, FALSE), c(1, 2), c(3, 4))
[1] 1 4
So ifelse is intended for the specific purpose of testing a vector of booleans and returning a vector of the same length, filled with elements taken from the (vector) yes and no arguments.
It is a common confusion, because of the function's name, to use this when really you want just a normal if () {} else {} construction instead.

I bet you want a simple if statement instead of ifelse - in R, if isn't just a control-flow structure, it can return a value:
> if(TRUE) c(1,2) else c(3,4)
[1] 1 2
> if(FALSE) c(1,2) else c(3,4)
[1] 3 4

Note that you can circumvent the problem if you assign the result inside the ifelse:
ifelse(TRUE, a <- c(1,2), a <- c(3,4))
a
# [1] 1 2
ifelse(FALSE, a <- c(1,2), a <- c(3,4))
a
# [1] 3 4

use `if`, e.g.
> `if`(T,1:3,2:4)
[1] 1 2 3

yeah, I think ifelse() is really designed for when you have a big long vector of tests and want to map each to one of two options. For example, I often do colors for plot() in this way:
plot(x,y, col = ifelse(x>2, 'red', 'blue'))
If you had a big long vector of tests but wanted pairs for outputs, you could use sapply() or plyr's llply() or something, perhaps.

Sometimes the user just needs a switch statement instead of an ifelse. In that case:
condition <- TRUE
switch(2-condition, c(1, 2), c(3, 4))
#### [1] 1 2
(which is another syntax option of Ken Williams's answer)

Here is an approach similar to that suggested by Cath, but it can work with existing pre-assigned vectors
It is based around using the get() like so:
a <- c(1,2)
b <- c(3,4)
get(ifelse(TRUE, "a", "b"))
# [1] 1 2

In your case, using if_else from dplyr would have been helpful: if_else is more strict than ifelse, and throws an error for your case:
library(dplyr)
if_else(TRUE,c(1,2),c(3,4))
#> `true` must be length 1 (length of `condition`), not 2

Found on everydropr:
ifelse(rep(TRUE, length(c(1,2))), c(1,2),c(3,4))
#>[1] 1 2
Can replicate the result of your condition to return the desired length

Why ifelse is not the same as if statement? [duplicate]

I've found R's ifelse statements to be pretty handy from time to time. For example:
ifelse(TRUE,1,2)
# [1] 1
ifelse(FALSE,1,2)
# [1] 2
But I'm somewhat confused by the following behavior.
ifelse(TRUE,c(1,2),c(3,4))
# [1] 1
ifelse(FALSE,c(1,2),c(3,4))
# [1] 3
Is this a design choice that's above my paygrade?

The documentation for ifelse states:
ifelse returns a value with the same
shape as test which is filled with
elements selected from either yes or
no depending on whether the element
of test is TRUE or FALSE.
Since you are passing test values of length 1, you are getting results of length 1. If you pass longer test vectors, you will get longer results:
> ifelse(c(TRUE, FALSE), c(1, 2), c(3, 4))
[1] 1 4
So ifelse is intended for the specific purpose of testing a vector of booleans and returning a vector of the same length, filled with elements taken from the (vector) yes and no arguments.
It is a common confusion, because of the function's name, to use this when really you want just a normal if () {} else {} construction instead.

I bet you want a simple if statement instead of ifelse - in R, if isn't just a control-flow structure, it can return a value:
> if(TRUE) c(1,2) else c(3,4)
[1] 1 2
> if(FALSE) c(1,2) else c(3,4)
[1] 3 4

Note that you can circumvent the problem if you assign the result inside the ifelse:
ifelse(TRUE, a <- c(1,2), a <- c(3,4))
a
# [1] 1 2
ifelse(FALSE, a <- c(1,2), a <- c(3,4))
a
# [1] 3 4

use `if`, e.g.
> `if`(T,1:3,2:4)
[1] 1 2 3

yeah, I think ifelse() is really designed for when you have a big long vector of tests and want to map each to one of two options. For example, I often do colors for plot() in this way:
plot(x,y, col = ifelse(x>2, 'red', 'blue'))
If you had a big long vector of tests but wanted pairs for outputs, you could use sapply() or plyr's llply() or something, perhaps.

Sometimes the user just needs a switch statement instead of an ifelse. In that case:
condition <- TRUE
switch(2-condition, c(1, 2), c(3, 4))
#### [1] 1 2
(which is another syntax option of Ken Williams's answer)

Here is an approach similar to that suggested by Cath, but it can work with existing pre-assigned vectors
It is based around using the get() like so:
a <- c(1,2)
b <- c(3,4)
get(ifelse(TRUE, "a", "b"))
# [1] 1 2

In your case, using if_else from dplyr would have been helpful: if_else is more strict than ifelse, and throws an error for your case:
library(dplyr)
if_else(TRUE,c(1,2),c(3,4))
#> `true` must be length 1 (length of `condition`), not 2

Found on everydropr:
ifelse(rep(TRUE, length(c(1,2))), c(1,2),c(3,4))
#>[1] 1 2
Can replicate the result of your condition to return the desired length

ifelse conditional assignent of tibbles [duplicate]

I've found R's ifelse statements to be pretty handy from time to time. For example:
ifelse(TRUE,1,2)
# [1] 1
ifelse(FALSE,1,2)
# [1] 2
But I'm somewhat confused by the following behavior.
ifelse(TRUE,c(1,2),c(3,4))
# [1] 1
ifelse(FALSE,c(1,2),c(3,4))
# [1] 3
Is this a design choice that's above my paygrade?

The documentation for ifelse states:
ifelse returns a value with the same
shape as test which is filled with
elements selected from either yes or
no depending on whether the element
of test is TRUE or FALSE.
Since you are passing test values of length 1, you are getting results of length 1. If you pass longer test vectors, you will get longer results:
> ifelse(c(TRUE, FALSE), c(1, 2), c(3, 4))
[1] 1 4
So ifelse is intended for the specific purpose of testing a vector of booleans and returning a vector of the same length, filled with elements taken from the (vector) yes and no arguments.
It is a common confusion, because of the function's name, to use this when really you want just a normal if () {} else {} construction instead.

I bet you want a simple if statement instead of ifelse - in R, if isn't just a control-flow structure, it can return a value:
> if(TRUE) c(1,2) else c(3,4)
[1] 1 2
> if(FALSE) c(1,2) else c(3,4)
[1] 3 4

Note that you can circumvent the problem if you assign the result inside the ifelse:
ifelse(TRUE, a <- c(1,2), a <- c(3,4))
a
# [1] 1 2
ifelse(FALSE, a <- c(1,2), a <- c(3,4))
a
# [1] 3 4

use `if`, e.g.
> `if`(T,1:3,2:4)
[1] 1 2 3

yeah, I think ifelse() is really designed for when you have a big long vector of tests and want to map each to one of two options. For example, I often do colors for plot() in this way:
plot(x,y, col = ifelse(x>2, 'red', 'blue'))
If you had a big long vector of tests but wanted pairs for outputs, you could use sapply() or plyr's llply() or something, perhaps.

Sometimes the user just needs a switch statement instead of an ifelse. In that case:
condition <- TRUE
switch(2-condition, c(1, 2), c(3, 4))
#### [1] 1 2
(which is another syntax option of Ken Williams's answer)

Here is an approach similar to that suggested by Cath, but it can work with existing pre-assigned vectors
It is based around using the get() like so:
a <- c(1,2)
b <- c(3,4)
get(ifelse(TRUE, "a", "b"))
# [1] 1 2

In your case, using if_else from dplyr would have been helpful: if_else is more strict than ifelse, and throws an error for your case:
library(dplyr)
if_else(TRUE,c(1,2),c(3,4))
#> `true` must be length 1 (length of `condition`), not 2

Found on everydropr:
ifelse(rep(TRUE, length(c(1,2))), c(1,2),c(3,4))
#>[1] 1 2
Can replicate the result of your condition to return the desired length

Using ifelse to change column names in R [duplicate]

I've found R's ifelse statements to be pretty handy from time to time. For example:
ifelse(TRUE,1,2)
# [1] 1
ifelse(FALSE,1,2)
# [1] 2
But I'm somewhat confused by the following behavior.
ifelse(TRUE,c(1,2),c(3,4))
# [1] 1
ifelse(FALSE,c(1,2),c(3,4))
# [1] 3
Is this a design choice that's above my paygrade?

The documentation for ifelse states:
ifelse returns a value with the same
shape as test which is filled with
elements selected from either yes or
no depending on whether the element
of test is TRUE or FALSE.
Since you are passing test values of length 1, you are getting results of length 1. If you pass longer test vectors, you will get longer results:
> ifelse(c(TRUE, FALSE), c(1, 2), c(3, 4))
[1] 1 4
So ifelse is intended for the specific purpose of testing a vector of booleans and returning a vector of the same length, filled with elements taken from the (vector) yes and no arguments.
It is a common confusion, because of the function's name, to use this when really you want just a normal if () {} else {} construction instead.

I bet you want a simple if statement instead of ifelse - in R, if isn't just a control-flow structure, it can return a value:
> if(TRUE) c(1,2) else c(3,4)
[1] 1 2
> if(FALSE) c(1,2) else c(3,4)
[1] 3 4

Note that you can circumvent the problem if you assign the result inside the ifelse:
ifelse(TRUE, a <- c(1,2), a <- c(3,4))
a
# [1] 1 2
ifelse(FALSE, a <- c(1,2), a <- c(3,4))
a
# [1] 3 4

use `if`, e.g.
> `if`(T,1:3,2:4)
[1] 1 2 3

yeah, I think ifelse() is really designed for when you have a big long vector of tests and want to map each to one of two options. For example, I often do colors for plot() in this way:
plot(x,y, col = ifelse(x>2, 'red', 'blue'))
If you had a big long vector of tests but wanted pairs for outputs, you could use sapply() or plyr's llply() or something, perhaps.

Sometimes the user just needs a switch statement instead of an ifelse. In that case:
condition <- TRUE
switch(2-condition, c(1, 2), c(3, 4))
#### [1] 1 2
(which is another syntax option of Ken Williams's answer)

Here is an approach similar to that suggested by Cath, but it can work with existing pre-assigned vectors
It is based around using the get() like so:
a <- c(1,2)
b <- c(3,4)
get(ifelse(TRUE, "a", "b"))
# [1] 1 2

In your case, using if_else from dplyr would have been helpful: if_else is more strict than ifelse, and throws an error for your case:
library(dplyr)
if_else(TRUE,c(1,2),c(3,4))
#> `true` must be length 1 (length of `condition`), not 2

Found on everydropr:
ifelse(rep(TRUE, length(c(1,2))), c(1,2),c(3,4))
#>[1] 1 2
Can replicate the result of your condition to return the desired length

How to drop element from named list without assignment?

Suppose I have a named list like
somelist <- list(a = 1, b = 5, c = 3)
I know that I can drop somelist$b, say, by assigning NULL to it:
somelist$b <- NULL
I suppose this is fine for interactive work, but not so much for programmatic work, because it forces the creation of otherwise superfluous variables.
For example, suppose that foo(42) evaluates to a list similar to somelist above, and that I want to pass the list resulting from dropping the b element from foo(42) to some other function bar. In this case, applying the method shown above would require the following:
superfluous.variable <- foo(42)
superfluous.variable$b <- NULL
bar(superfluous.variable)
rm(superfluous.variable)
I'm looking for a way to pass to bar the modified results from foo that does not require these superfluous assignments. The four lines above would collapse to a single line:
bar(drop.item.from.list(foo(42), item.to.drop = "b"))
Does R already have something like the hypothetical drop.item.from.list function above?

You can do that removal on the fly with replace()
replace(somelist, "b", NULL)
# $a
# [1] 1
#
# $c
# [1] 3
It works for multiple variables as well ...
replace(somelist, c("a", "b"), NULL)
# $c
# [1] 3
So just wrap that in bar() and the original list remains intact.
Note: I am not exactly sure what you are doing with foo(42) but you state that the resulting list takes a similar structure, so this should be fine for that.

We can try with setdiff
bar(foo(42)[setdiff(names(somelist), "b")])
as the setdiff subsets the 'somelist'
somelist[setdiff(names(somelist), "b")]
#$a
#[1] 1
#$c
#[1] 3
We can also use this to subset for multiple variables
somelist[setdiff(names(somelist), c("a", "b"))]
#$c
#[1] 3

Resources