Implement function callbacks to make function modifications uniform and streamlined about cobrexa.jl

[laurentheirendt]

closing

[laurentheirendt]

Do you have an example? I am missing somehow the context here :/

[stelmo]

Yes! So I think it is pretty important to allow users to modify analysis functions in a uniform way. Currently fba, pfba and fva have the following keyword arguments: objective_func, weights, solver_attributes, constraints. These keyword arguments are then used inside the functions to modify the problems, e.g. the user can specify (a) new objective function(s) with optional weights (if there are multiple fluxes that need to be summed), etc. The issue is that this leads to lots of duplicate code, e.g. lines 44-90 in fba are basically duplicated in pfba and fva since the modifications one would typically make are usually similar in form.

@exaexa suggested that we use callbacks to streamline all of this and gave me the following example:

function fba(...; model_mod = (m, om) -> nothing)
    om = makeOptimizationModel(...)
    model_mod(m, om) 
    optimize(om)
    return(...(om))
end

modify_weights(r::Reaction) =
    (m, om) -> begin
        ...
        @objective(om, sense, sum(opt_weights[i] * v[i] for i in objective_indices))
    end

fba(model, model_mod = [[modify_weight("ATPw"), remove_reaction("asasd")]] )

Or something similar to this?

[laurentheirendt]

I like the idea 👍🏼 Do you think this has any effect on performance?

[exaexa]

effect on performance

negligible, except perhaps for the first-time compile

[laurentheirendt]

okay cool, then go ahead 😄

[stelmo]

Is being addressed by #53