Replace eval with macros

[GiggleLiu]

The minimum fix should be like

using JunctionTrees

problem_number = "34"
problem_filename = joinpath("Promedus_" * problem_number)
problem_dir = joinpath(@__DIR__, "../examples/problems/Promedus/", problem_number)
uai_filepath = joinpath(problem_dir, problem_filename * ".uai")
uai_evid_filepath = joinpath(problem_dir, problem_filename * ".uai.evid")
uai_mar_filepath = joinpath(problem_dir, problem_filename * ".uai.MAR")
td_filepath = joinpath(problem_dir, problem_filename * ".tamaki.td")

reference_marginals = JunctionTrees.read_uai_mar_file(uai_mar_filepath)
obsvars, obsvals = JunctionTrees.read_uai_evid_file(uai_evid_filepath)

# ------------------------------------------------------------------------------
# Posterior marginals given evidence
# ------------------------------------------------------------------------------

macro inferencemodel(uai_filepath, uai_evid_filepath)
    compile_algo(
            eval(uai_filepath),
            uai_evid_filepath = eval(uai_evid_filepath),
        )
end

algo = @inferencemodel uai_filepath uai_evid_filepath

marginals = algo(obsvars, obsvals) |> x -> map(y -> y.vals, x)

[mroavi]

Would there be any advantage of using a non-standard string literal over a macro?
E.g.:

algo = compile"problems/asia/asia.uai"
obsvars, obsvals = Int64[], Int64[]
marginals = run_algo(algo, obsvars, obsvals)

[GiggleLiu]

Would there be any advantage of using a non-standard string literal over a macro? E.g.:

algo = compile"problems/asia/asia.uai"
obsvars, obsvals = Int64[], Int64[]
marginals = run_algo(algo, obsvars, obsvals)

It looks good. The problem is how to include also uai_evid_filepath.

[mroavi]

Right. You have a point. But putting aside the extra arguments that we might want to pass to compile_algo, would this bring benefits over a standard macro?
I think we should go to the approach you propose then. I have a question about macros though. compile_algo can take a series of optional parameters that customize the generation of the algorithm. Do macros support optional arguments as well? I couldn't find anything related to this in the docs section about metaprogramming.

[GiggleLiu]

Right. You have a point. But putting aside the extra arguments that we might want to pass to compile_algo, would this bring benefits over a standard macro? I think we should go to the approach you propose then. I have a question about macros though. compile_algo can take a series of optional parameters that customize the generation of the algorithm. Do macros support optional arguments as well? I couldn't find anything related to this in the docs section about metaprogramming.

It is easy to do, just create a macro like the following

julia> using MLStyle

julia> macro m(ex1, kwargs...)
    d = Dict()
    for kw in kwargs
        @match kw begin
            :($a = $b) => begin
                d[a] = b
            end
        end
    end
    d
end
@m (macro with 1 method)

julia> @m a b=3 c=5
Dict{Any, Any} with 2 entries:
  :b => 3
  :c => 5

[mroavi]

Ahh, macros do support varargs. Great. Then I think we should go with this approach. I also want to rename the run_algo function to infer.

[mroavi]

The minimum fix should be like

using JunctionTrees

problem_number = "34"
problem_filename = joinpath("Promedus_" * problem_number)
problem_dir = joinpath(@__DIR__, "../examples/problems/Promedus/", problem_number)
uai_filepath = joinpath(problem_dir, problem_filename * ".uai")
uai_evid_filepath = joinpath(problem_dir, problem_filename * ".uai.evid")
uai_mar_filepath = joinpath(problem_dir, problem_filename * ".uai.MAR")
td_filepath = joinpath(problem_dir, problem_filename * ".tamaki.td")

reference_marginals = JunctionTrees.read_uai_mar_file(uai_mar_filepath)
obsvars, obsvals = JunctionTrees.read_uai_evid_file(uai_evid_filepath)

# ------------------------------------------------------------------------------
# Posterior marginals given evidence
# ------------------------------------------------------------------------------

macro inferencemodel(uai_filepath, uai_evid_filepath)
    compile_algo(
            eval(uai_filepath),
            uai_evid_filepath = eval(uai_evid_filepath),
        )
end

algo = @inferencemodel uai_filepath uai_evid_filepath

marginals = algo(obsvars, obsvals) |> x -> map(y -> y.vals, x)

There is an issue with how the macro evaluates the uai_filepath. At first, I thought it was just a problem with the macro's hygiene, but it seems to be more complex than that. The macro is limited to only the expression that it's given, and without using eval (which is a global construct and doesn't work in, for example, let or @testset blocks), it cannot access the variable values in that expression.

One approach to solve this is to pass the problem name to the macro as a String literal, e.g. "Promedus_14", and then move the lines

uai_filepath = joinpath(problem_dir, problem * ".uai")
...

But this would have a negative impact on the interface for constructing the posterior marginals algorithms, as it would be unable to handle variables with bounded values.