Metaprogramming: Nested function variable interpolation

General Usage
,
1

Hello,

I’m working on a piece of code that generates a math expression with embedded variables, and I want to generate a function that will take those variables as inputs and evaluate the math expression using the given inputs. E.g., if my math expression ends up being a + b, I want to have a generated function that takes a and b as inputs and will evaluate a + b. The purpose of this is that generating my math expression takes a lot of time, but once I have the math expression, evaluating it should be fast, and I want to be able to evaluate it with different variables a large number of times.

My thinking was to use a nested function approach where the variable names and final math expression are interpolated into an inner function definition, which hopefully I can later call to evaluate the equation. Here’s a contrived example of what I’m trying to do where the math expression is just a sum of num variables:

function make_func(num::Int)

    # Generate some variables
    vars = String[]
    for i = 1:num
        push!(vars, "v"*string(i))
    end

    # Generate an equation using these variables
    my_sum = ""
    for var in vars
        my_sum = my_sum * var * " + "
    end
    my_sum = my_sum[1:end-3]

    # Create a function that will evaluate this equation
    @eval begin
        function new_func(vars...)
            the_sum = Meta.parse($my_sum)
            return the_sum
        end
    end
end

Ideally, running make_func(3) would generate a function called new_func which takes 3 inputs, calculates the sum of the values, and returns the sum. But this doesn’t work:

julia> make_func(3); new_func(1,5,7)
:(v1 + v2 + v3)

julia> new_func()
:(v1 + v2 + v3)

For one thing, new_func(vars...) is clearly not creating a function where the elements of “vars” are the arguments. And for another thing, the :(v1 + v2 + v3) expression seems to be detached from the inputs to new_func, which means I can’t get its evaluation.

Is there any way to do something like this? Some fix to the nested function? A way to evaluate a code-generated expression multiple times without re-generating it?

Any help would be greatly appreciated!

Thanks!

2

You could just use Symbolics.jl

using Symbolics

function make_func(num::Int)
    # Generate some variables
    @variables v[1:num]
    v = collect(v)
    ex = sum(v)
    f_oop, _= build_function([ex], v..., expression = Val{false})
    f_oop
end

f_ = make_func(3)
f_(1,2,3)
1 Like