InPartS.jl - Agents.jl interop

[JonasIsensee]

Hi,
I tried to implement the API glue for InPartS.jl with Agents.jl.
I made some progress but also encountered the following things:

• Plotting seems to rely on agent.pos to give the position. InPartS uses :centerpos. So I had to implement a wrapper type.
  I (of course) need to implement a Model wrapper type as well. This seems to need to have be of the form InPartSModel{Agents.ContinuousSpace{2}} <: Agents.AgentBasedModel{ContinuousSpace{2}}

• I don't understand how time works in Agents. The data-collection for abmexploration appears to implicitly require time::Int (?) unless it is event-based but conceptually our models just have continuous time (adaptive time steppers)
  As a consequence, what should I do to add something like the time stepping API?

• side remark: In InPartS we found that is way more efficient if we provide self-generated meshes to Makie (rather than providing polygons)

• In InPartS, there is by default, no mutable global set of parameters. I'm wondering if that defeats most of the purpose of the interactive tools.

Anyway, Here"s my WIP

using InPartS, Agents
using InPartSBiome
using GLMakie, InPartSMakie
using Random

## Agents Model APINothing

mutable struct InPartSModelv4{S} <: Agents.AgentBasedModel{S}
    sim::InPartS.Simulation
end
InPartSModel = InPartSModelv4{Agents.ContinuousSpace{2}}


mutable struct AgentWrapper <: Agents.AbstractAgent
    cell::InPartS.AbstractParticle
end
function Base.getproperty(a::AgentWrapper, s::Symbol)
    cell = getfield(a, :cell)
    if s == :pos
        return cell.centerpos
    end
    getproperty(cell, s)
end
Agents.abmproperties(model::InPartSModel) = nothing

Agents.abmrng(model::InPartSModel) = model.sim.rng

function Agents.abmspace(model::InPartSModel) 
    domain =  getfield(model, :sim).domain
    ContinuousSpace(domainsize(domain)) 
end
Agents.abmtime(model::InPartSModel) = getfield(model, :sim).step

Agents.agent_container(model::InPartSModel) = AgentWrapper.(particles(getfield(model, :sim)))

Agents.agents_space_dimensionality(::Domain2D) = 2

Agents.agenttype(model::InPartSModel) = AgentWrapper#InPartS.particletype(getfield(model, :sim))

#nextid(model::InPartSModel) = model.sim.nextid
function Agents.add_agent_to_model!(model::InPartSModel, agent)
    push!(sim._addition_buffer, agent)
    InPartS.flushadditions!(sim)
    return agent.id
end

function Agents.remove_agent_from_model!(agent, model::InPartSModel)
    push!(sim._deletion_buffer, agent.id)
    InPartS.flushdeletions!(sim)
end

mod = Base.get_extension(Agents, :AgentsVisualizations)
mod._default_dts_from_model(::InPartSModel) = 0.01:0.01:0.1

using Statistics: mean
black(a) = a.growthrate < 1.0
white(a) = a.growthrate >= 1.0
adata = [(black, count), (white, count)]
temperature(model) = length(particles(getfield(model,:sim)))#mean(model.temperature)
mdata = [temperature]
fig, abmobs = abmexploration(ipsmodel;
    agent_color=:blue,
    agent_size = 20,
    agent_marker = 'o',
    agentsplotkwargs = (; strokewidth = 1.0,),
    adata, alabels = ["Black daisys", "White daisys"],
    mdata, mlabels = ["T"]
)


##

sim = Simulation(
    SimpleParticleContainer{2, RodCell{AreaMobility}}();
    domainsize = (30, 15),
    boundaries = (PeriodicBoundary, Boundary),
    rng = Random.Xoshiro(42)
)

addparticle!(sim;
    centerpos = SA[20.0, 10.0],
    φ = π*rand(sim.rng),
    params = RodCellParams(maxlength = 3, youngsmodulus = 1e4)
)
InPartS.flushadditions!(sim)
evolve!(sim, 1.0, tss = SimpleAdaptiveStepper(0.01))


ipsmodel = InPartSModel(sim)


fig, ax, abmobs = abmplot(ipsmodel; add_controls = true, agent_marker='o')
fig

# integration using InPartS
dxmax = 0.005
dtmax = 0.01
evolve!(sim, 0.1; tss = SimpleAdaptiveStepper(dxmax, dtmax = dtmax))

# Regular InPartSMakie plot for reference
fig = Figure();
ax = Axis(fig[1,1], aspect = DataAspect())
plt = plot!(ax, sim, particles = (;colorfunction = InPartSBiome.orientationcolor))
fig