allow [:a,:b,:c,:d] => fun => new_column_name syntax

[floswald]

I was trying to do this

using DataFrames
julia> df = DataFrame(a=1:3, b=4:6,c = rand(3))

# expected
julia> select(df, [:a,:b] => (a,b) -> a./b)
3×1 DataFrame
│ Row │ a_b_function │
│     │ Float64      │
├─────┼──────────────┤
│ 1   │ 0.25         │
│ 2   │ 0.4          │
│ 3   │ 0.5          │

# not expected!
julia> select(df, [:a,:b] => (a,b) -> a./b => :d)
3×1 DataFrame
│ Row │ a_b_function         │
│     │ Pair…                │
├─────┼──────────────────────┤
│ 1   │ [0.25, 0.4, 0.5]=>:d │
│ 2   │ [0.25, 0.4, 0.5]=>:d │
│ 3   │ [0.25, 0.4, 0.5]=>:d │

# not expected either
julia> select(df, [:a, :b] => (a,b) -> a ./ b => "new")
3×1 DataFrame
│ Row │ a_b_function            │
│     │ Pair…                   │
├─────┼─────────────────────────┤
│ 1   │ [0.25, 0.4, 0.5]=>"new" │
│ 2   │ [0.25, 0.4, 0.5]=>"new" │
│ 3   │ [0.25, 0.4, 0.5]=>"new" │

It seems that given you allow that syntax when transforming a single column, why not also on multiple ones? the problem with the current approach? well, if I want to add two transformed columns which get the same generated name, I can't do that in one go. Or what's the best approach here?

julia> select(df, [:a, :b] => (a,b) -> a ./ b, [:a, :b] => (a,b) -> a .+ b)
ERROR: ArgumentError: duplicate target column name a_b_function passed

[bkamins]

This is the problem with Julia Base not with DataFrames.jl.

Note the difference:

julia> x -> x => :new
#5 (generic function with 1 method)

julia> (x -> x) => :new
var"#7#8"() => :new

So, as you can see, you were struck by the operator precedence issue, and you have to wrap the anonymous function in ( and ) like this:

julia> select(df, [:a,:b] => ((a,b) -> a./b) => :d)
3×1 DataFrame
│ Row │ d       │
│     │ Float64 │
├─────┼─────────┤
│ 1   │ 0.25    │
│ 2   │ 0.4     │
│ 3   │ 0.5     │

julia> select(df, [:a, :b] => ((a,b) -> a ./ b) => "new")
3×1 DataFrame
│ Row │ new     │
│     │ Float64 │
├─────┼─────────┤
│ 1   │ 0.25    │
│ 2   │ 0.4     │
│ 3   │ 0.5     │

[floswald]

cool! thanks!