Accessing the data should not be time-consuming at all. TypedTables shouldn’t be an improvement over DataFrames for data access alone.
It’s likely the real problem is that you are not using a function barrier. For best performance, write functions that act on the vectors directly, not the whole data frame.
function foo(a, b)
a[100] + b[101]
end
foo(df.a, df.b)
DataFramesMeta.jl’s @with macro can help with this.
@with df begin
:a[100] + :b[101]
end