Conditionally setting a column in dataframes.jl

demo.jl

using DataFrames, BenchmarkTools, DataFramesMeta
eg_df(n=100_000) = DataFrame(a=rand(1:10, n), b=rand('a':'z', n), c=rand('A':'Z', n))


# for eachrow loop
# 7.132 ms (239156 allocations: 4.10 MiB)
@btime let
    df = $(eg_df())
    for row in eachrow(df)
        if row.a == 1
            row.b = row.c
        end
    end
end;

# masking
# 112.436 μs (25 allocations: 117.98 KiB)
@btime let
    df = $(eg_df())
    df[df.a .== 1, :b] .== df[df.a .== 1, :c]
end

# For indexed
# 6.678 ms (249279 allocations: 3.96 MiB)
@btime let
    df = $(eg_df())
    for ii in 1:nrow(df)
        if df[ii, :a] == 1
            df[ii, :b] = df[ii, :c]
        end
    end
end;

# For indexed inbounds
# 4.335 ms (139204 allocations: 2.28 MiB)
@btime let
    df = $(eg_df())
    @inbounds for ii in 1:nrow(df)
        if df[ii, :a] == 1
            df[ii, :b] = df[ii, :c]
        end
    end
    df
end;


# Make a function
#  22.892 μs (0 allocations: 0 bytes)
function set_b!(a, b, c)
    @inbounds for ii in eachindex(a)
        if a[ii] == 1
            b[ii] = c[ii]
        end
    end
end
@btime let
    df = $(eg_df())
    set_b!(df.a, df.b, df.c)
end;


# byrow!
#  248.458 μs (24 allocations: 1.53 MiB)
@btime let
    df = $(eg_df())
    @byrow! df begin 
        :b = :a == 1 ? :b : :c 
    end
end;


# @transform
#  109.362 μs (26 allocations: 1.91 MiB)
@btime let
    df = $(eg_df())
    @transform(df, b = ifelse.(:a .== 1, :b, :c))
end;

# transform ternery
#   103.689 μs (97 allocations: 1.53 MiB)
@btime let
    df = $(eg_df())
    transform(df, [:a, :b, :c] => ByRow((a, b,c) -> a == 1 ? b : c) => :b)
end;

# transform ifelse
# 100.875 μs (83 allocations: 1.53 MiB)
@btime let
    df = $(eg_df())
    transform(df, [:a, :b, :c] => ByRow((a, b,c) -> ifelse(a == 1, b, c)) => :b)
end;

summary.jl

# Make a function
#     23 μs (0 allocations: 0 bytes)

# transform ifelse
#    101 μs (83 allocations: 1.53 MiB)
# transform ternery
#    104 μs (97 allocations: 1.53 MiB)
# @transform
#    109 μs (26 allocations: 1.91 MiB)
# masking
#    112 μs (25 allocations: 117.98 KiB)
# byrow!
#    248 μs (24 allocations: 1.53 MiB)

# For indexed inbounds
#   4335 μs (139204 allocations: 2.28 MiB)
# For indexed
#   6678 μs (249279 allocations: 3.96 MiB)
# for eachrow loop
#   7132 μs (239156 allocations: 4.10 MiB)