Can you avoid stating the arguments in an anonymous function?

Is it possible to avoid stating the arguments in an anonymous function?

x = (a=1,b=2)
( (a,b) -> a+b )(x...)

works fine, but could you do something like
( (_...) -> a+b )(x...)

x would have to contain elements a and b. Otherwise there would be a run-time error.

I think you can only do that with varags, and it’s not exactly what you want but

x = (a...) -> sum(a)
julia> x(1,2,3)

what does this mean? (i.e. what do you want it to do?) For example, what even is a,b? If you want to do something and don’t care about the argument, you should

julia> x
(a = 1, b = 2)

julia> values(x)
(1, 2)

julia> sum(x)

I’d like to use variables in the body of a function without listing them explicitly as function inputs.

(probably not possible. just thought I’d check)

how is that logically possible? how does the function know what variables to use then? maybe you’re thinking about closure?

julia> g(y0) = x-> x+y0
g (generic function with 1 method)

julia> g(3)(4)

Background –
For DataFrames, I’d like an easy way to do row-level operations, and use the column names as variable names and have any created variables added automatically as new columns.

in this discussing a nice method was suggested

but it requires writing a function where you have to specify the column names as inputs and outputs.

function f(; a, b, ... )
    c = a + b
    d = a - b
    (; a, b, c, d) 

D = DataFrame( a=[1,2], b=[3.4) )

@chain D begin
       transform( AsTable(:) =>   ByRow(x->f(x...)) => AsTable )

I’d like to avoid re-specifying the column names at all. Something like:

@chain D begin
        ***drop to row level****   begin
            c = a + b
            d = a -  b

Yes. DataFramesMeta.jl does exactly this.

julia> df = DataFrame(a = [1, 2, 3], b = [5, 6, 7]);

julia> @rtransform df begin
           :c = :a + :b
           :d = :a - :b
3×4 DataFrame
 Row │ a      b      c      d
     │ Int64  Int64  Int64  Int64
   1 │     1      5      6     -4
   2 │     2      6      8     -4
   3 │     3      7     10     -4

I have DataFramesMeta installed but for some reason @rtransform isn’t recognised

Check your version. Current version is 0.9.1 @rtransform should be in 0.9.0 and newer.

1 Like

works :slight_smile: thank you so much.

is there any way to avoid the colons in the variable names?

Related question - there are various DF filtering functions @where, @subset
All seem to require vectorising the conditional statement eg :a .> 3.

Is there a way of doing @rsubset so you don’t have to vectorise the statument e.g :a > 3

No, you can’t avoid the colons. Columns are referenced as :x.

The reason for this is that we need a way to distinguish, at parse time, the columns in a data frame from other variables. Without knowing anything about the data frame.

x = 1
@rtransform df :y = :x + x

Obviously it’s possible to make unquoted symbols, i.e. x column references and leave special syntax for everything else. But then you would hvae to apply lots of escaping rules.

@rtransform df y = begin 
    $x = 100
    x + z

This might get out of hand when people want to use missing, map with a function as the first argument, etc.

I would put a positive spin on this and say the use of :x makes code more readable because you can distinguish easily between columns and variables, which can get confusing in dplyr.

Finally, @rsubset exists in 0.9.0 and newer. @where is deprecated in favor of subset, so there is just one filtering function in DataFramesMeta.jl.


@rsubset Awesome thank you :slight_smile:

Last question. Within an @rtransform bloc new columns can’t depend on other new columns.
Is there a way to have a row level block with multiple interdependent new columns?

Yes! I’m releasing a new version tomorrow with this feature. Stay tuned for the announcement.


Had some more questions. Put them in a new topic with appropriate heading.

1 Like