It's all about graphs, right?

[mauro-milella]

Hi guys,

Do you see any differences between graph theory and modal symbolic learning?
I don't, maybe I am being quite superficial?

I am writing this issue to anticipate the refactoring/simplification work are going to do in the next months.

This is a fragment of code I just wrote in ModalAssociationRules.jl, to fit my experiments with generic AbstractDataset types rather than UniformFullDimensionalDatasets. It is inspired by some code I read from @alberto-paparella.

Do you think this is garbage? The underlying philosophy is: please, gimme my graph, gimme my coloured edges, and let me fix which symbol holds on every node.

No jungle types. Stick to Graphs.jl as much as possible, and keep it simple. No strongly-coupled code enabling heavy optimizations.

struct Logiset <: SoleData.AbstractLogiset
    instances::Vector{KripkeStructure}
end

After the above, we could implement the standard Vector interface via @forward macro (Lazy.jl package). Something like:

@forward Logiset.instances Logiset begin
    length
    getindex
    setindex!
    push!
    pop!
    iterate
end

The rest could be something like:

# this is not the complete interface, but only a few dispatches I needed in ModalAssociationRules.jl

function instances(logiset::Logiset)
    return logiset.instances
end

function ninstances(logiset::Logiset)
    return logiset |> length # and this is a renaming for length(::Vector{KripkeStructure}), because of @forward
end

function getinstance(
    logiset::Logiset,
    i::Int64
)::SoleLogics.LogicalInstance
    return SoleLogics.LogicalInstance(
        SoleLogics.InterpretationVector(logiset |> instances), i)
end

function frame(logiset::Logiset, i::Int64)
    instances(logiset)[i] |> frame
end

[eduardstan]

Dear Mauro,

I understand your concern, but could you provide more context? As I recall, in Graph Learning, you can infer, for example, new edges in the input graph (e.g., a potential collaboration between researchers who haven’t collaborated before).

Best

[PasoStudio73]

I see the point of this issue simply because @mauro-milella and me have talked several times about this.
My approval. Let's see if this idea leads us in some improvement.

[alberto-paparella]

Hi all,

I believe this must be addressed and designed carefully, but I'm totally on board!

We all know that debugging SoleData.jl can be quite frustrating sometime, and I'm sure the whole type hierarchy can be simplified a lot leveraging Graphs.jl as much as possible.

@eduardstan I'm pretty sure @mauro-milella meant graph theory and graphs in general, NOT graph learning (;
Editing the issue appropriately.

Cheers

[Perro2110]

I agree with all of this, it's fine by me! :)

[eduardstan]

Ok, it makes sense now. I also believe that some aspects of SoleData could benefit from updates, considering that the system has been in use for several years and might be experiencing data obsolescence or increased inaccuracies. All in all, I agree that refreshing the crucial data matters is essential for maintaining accurate and reliable information within our operations.