Fix of type instability in SIR model:

- Real => Float64 for better performance (Real is an abstract type)
- Should not create a type inside the model because the compiler fails
at estimating the type of the output of the f function (detected with
@code_warntype)
- Intentionaly ER is not as well improved as SIR in order to check the
differences of performance.

core/model.jl

@@ -12,19 +12,19 @@ mutable struct CTMC <: ContinuousTimeModel
     _map_obs_var_idx::Dict # maps str to observed state space
     map_param_idx::Dict # maps str in parameter space
     l_name_transitions::AbstractVector{String}
-    p::AbstractVector{Real}
+    p::AbstractVector{Float64}
     x0::AbstractVector{Int}
-    t0::Real
+    t0::Float64
     f::Function
     g::AbstractVector{String} # of dimension dobs
-    _g_idx::AbstractVector{Int} # of dimension dobs
+    _g_idx::Vector{Int} # of dimension dobs
     is_absorbing::Function
-    time_bound::Real
+    time_bound::Float64
 end
 
 function CTMC(d::Int, k::Int, map_var_idx::Dict, map_param_idx::Dict, l_name_transitions::AbstractVector{String}, 
-              p::AbstractVector{Real}, x0::AbstractVector{Int}, t0::Real, 
-              f::Function, is_absorbing::Function; g::AbstractVector{String} = keys(map_var_idx), time_bound::Real = Inf)
+              p::AbstractVector{Float64}, x0::AbstractVector{Int}, t0::Float64, 
+              f::Function, is_absorbing::Function; g::AbstractVector{String} = keys(map_var_idx), time_bound::Float64 = Inf)
     dobs = length(g)
     _map_obs_var_idx = Dict()
     _g_idx = Vector{Int}(undef, dobs)
@@ -72,9 +72,9 @@ end
 
 is_bounded(m::Model) = m.time_bound < Inf
 function check_consistency(m::Model) end
-function simulate(m::Model, n::Int; bound::Real = Inf)::AbstractObservations end
-function set_param!(m::Model, p::AbstractVector{Real})::Nothing end
-function get_param(m::Model)::AbstractVector{Real} end
+function simulate(m::Model, n::Int; bound::Float64 = Inf)::AbstractObservations end
+function set_param!(m::Model, p::AbstractVector{Float64})::Nothing end
+function get_param(m::Model)::AbstractVector{Float64} end
 
 function load_model(name_model::String)
     include(pathof(@__MODULE__) * "/../../models/" * name_model * ".jl")

core/observations.jl

@@ -4,14 +4,14 @@ ContinuousObservations = AbstractVector{AbstractTrajectory}
 
 struct Trajectory <: AbstractTrajectory
     m::ContinuousTimeModel
-    values::AbstractMatrix{Real}
-    times::AbstractVector{Real}
+    values::AbstractMatrix{Float64}
+    times::AbstractVector{Float64}
     transitions::AbstractVector{Union{String,Nothing}}
 end
 
 function +(σ1::AbstractTrajectory,σ2::AbstractTrajectory) end
 function -(σ1::AbstractTrajectory,σ2::AbstractTrajectory) end
-function δ(σ1::AbstractTrajectory,t::Real) end
+function δ(σ1::AbstractTrajectory,t::Float64) end
 function get_obs_variables(σ::AbstractTrajectory) end
 
 get_values(σ::AbstractTrajectory, var::String) = 

models/ER.jl

@@ -3,7 +3,7 @@
 import StaticArrays: SVector, SMatrix, @SMatrix
 
 State = SVector{4, Int}
-Parameters = SVector{3, Real}
+Parameters = SVector{3, Float64}
 
 d=4
 k=3
@@ -14,7 +14,7 @@ p = Parameters(1.0, 1.0, 1.0)
 x0 = State(100, 100, 0, 0)
 t0 = 0.0
 
-function f(xn::State, tn::Real, p::Parameters)
+function f(xn::State, tn::Float64, p::Parameters)
     a1 = p[1] * xn[1] * xn[2]
     a2 = p[2] * xn[3]
     a3 = p[3] * xn[3]
@@ -45,10 +45,10 @@ function f(xn::State, tn::Real, p::Parameters)
     
     return xnplus1, tnplus1, transition
 end
-is_absorbing_sir(p::Parameters,xn::State) = 
+is_absorbing_er(p::Parameters,xn::State) = 
     (p[1]*xn[1]*xn[2] + (p[2]+p[3])*xn[3]) == 0.0
 g = SVector("P")
 
-ER = CTMC(d,k,dict_var,dict_p,l_tr,p,x0,t0,f,is_absorbing_sir; g=g)
+ER = CTMC(d,k,dict_var,dict_p,l_tr,p,x0,t0,f,is_absorbing_er; g=g)
 export ER
 

models/SIR.jl

 
 import StaticArrays: SVector, SMatrix, @SMatrix
 
-State = SVector{3, Int}
-Parameters = SVector{2, Real}
-
 d=3
 k=2
 dict_var = Dict("S" => 1, "I" => 2, "R" => 3)
 dict_p = Dict("ki" => 1, "kr" => 2)
 l_tr = ["R1","R2"]
-p = Parameters(0.0012, 0.05)
-x0 = State(95, 5, 0)
+p = SVector(0.0012, 0.05)
+x0 = SVector(95, 5, 0)
 t0 = 0.0
-function f(xn::State, tn::Real, p::Parameters)
+function f(xn::SVector{3, Int}, tn::Float64, p::SVector{2, Float64})
     a1 = p[1] * xn[1] * xn[2]
     a2 = p[2] * xn[2]
     l_a = SVector(a1, a2)
@@ -37,13 +34,13 @@ function f(xn::State, tn::Real, p::Parameters)
     end
  
     nu = l_nu[:,reaction]
-    xnplus1 = State(xn[1]+nu[1], xn[2]+nu[2], xn[3]+nu[3])
+    xnplus1 = SVector(xn[1]+nu[1], xn[2]+nu[2], xn[3]+nu[3])
     tnplus1 = tn + tau
     transition = "R$(reaction)"
 
     return xnplus1, tnplus1, transition
 end
-is_absorbing_sir(p::Parameters,xn::State) = (p[1]*xn[1]*xn[2] + p[2]*xn[2]) == 0.0
+is_absorbing_sir(p::SVector{2, Float64}, xn::SVector{3, Int}) = (p[1]*xn[1]*xn[2] + p[2]*xn[2]) == 0.0
 g = SVector("I")
 
 SIR = CTMC(d,k,dict_var,dict_p,l_tr,p,x0,t0,f,is_absorbing_sir; g=g)