How to convert all nothings in dataframe to missing?

General Usage
2

From the point of view of the function itself, I do not know if you can get faster than the simple loop:

julia> function nothing_to_missing!(x)
         for i in eachindex(x)
           if isnothing(x[i])
             x[i] = missing
           end
         end
       end

However, the performance is very much dependent on the type of vector:

# Vector Any[]
julia> @btime nothing_to_missing!(x) setup=(x=Any[isodd(i) ? nothing : 1 for i in 1:1000 ]) evals=1
  4.905 μs (0 allocations: 0 bytes)

# Vector Union{Int,Nothing,Missing}
julia> @btime nothing_to_missing!(x) setup=(x=Union{Int,Nothing,Missing}[isodd(i) ? nothing : 1 for i in 1:1000 ]) evals=1
  1.343 μs (0 allocations: 0 bytes)

This for arrays. I do not know if dataframes have some specific behavior concerning these values.

Edit: With a DataFrame it seems to be much slower. But I am not completely sure if this benchmark makes sense, I am not a regular user of data frames:

julia> function nothing_to_missing!(df,col)
         for i in eachindex(df[col])
           if isnothing(df[col][i])
             df[col][i] = missing
           end
         end
       end

julia> @btime nothing_to_missing!(df,1) setup=(df=DataFrame([Any[isodd(i) ? nothing : 1 for i in 1:1000 ]],[:x])) evals=1
  145.723 μs (1978 allocations: 46.52 KiB)

julia> @btime nothing_to_missing!(df,1) setup=(df=DataFrame([Union{Int,Missing,Nothing}[isodd(i) ? nothing : 1 for i in 1:1000 ]],[:x])) evals=1
  127.832 μs (1978 allocations: 46.52 KiB)
1 Like