From 189c2fa4d2aea281981454c206957a8bf6417abc Mon Sep 17 00:00:00 2001
From: pjentsch <pjentsch@uwaterloo.ca>
Date: Wed, 21 Apr 2021 22:16:47 -0400
Subject: [PATCH] improve performance in model loading
---
CovidAlertVaccinationModel/src/ABM/abm.jl | 4 +--
CovidAlertVaccinationModel/src/ABM/agents.jl | 37 +++++++++-----------
CovidAlertVaccinationModel/src/ABM/solve.jl | 18 ++++++++--
CovidAlertVaccinationModel/src/data.jl | 16 ++++-----
4 files changed, 42 insertions(+), 33 deletions(-)
diff --git a/CovidAlertVaccinationModel/src/ABM/abm.jl b/CovidAlertVaccinationModel/src/ABM/abm.jl
index 687088d..7b78faf 100644
--- a/CovidAlertVaccinationModel/src/ABM/abm.jl
+++ b/CovidAlertVaccinationModel/src/ABM/abm.jl
@@ -2,7 +2,7 @@
const population = 14.57e6 #population of ontario
const workschool_mixing, rest_mixing = load_mixing_matrices()
const age_bins = [(0.0, 25.0),(25.0,65.0),(65.0,Inf)]
-
+const household_data = read_household_data()
default(dpi = 300)
default(framestyle = :box)
@@ -13,7 +13,7 @@ Runs the model with 5k households for 500 timesteps.
"""
function bench()
Random.seed!(RNG,1)
- steps = 500
+ steps = 50
model_sol = ModelSolution(steps,get_parameters(),5000)
recording = DebugRecorder(steps)
output = solve!(model_sol,recording )
diff --git a/CovidAlertVaccinationModel/src/ABM/agents.jl b/CovidAlertVaccinationModel/src/ABM/agents.jl
index 5b830aa..81f0937 100644
--- a/CovidAlertVaccinationModel/src/ABM/agents.jl
+++ b/CovidAlertVaccinationModel/src/ABM/agents.jl
@@ -26,23 +26,29 @@ end
Generates a complete graph from a vector of household compositions. Each household composition is a 3 element vectors (one for each demographic group) of integers where each element describes the number of the corresponding demographic group present in that household.
This function wires together a graph such that each household is in a complete subgraph. It is much faster than the `mapreduce` solution used previously.
+
+Also returns a vector of AgentDemographic representing each agent, defined by the household compositions.
"""
-function complete_graph_from_households_composition(households_composition)
+@views function complete_graph_from_households_composition(households_composition)
total_household_pop = sum(sum.(households_composition))
+ population_list = Vector{AgentDemographic}(undef,total_household_pop)
network = SimpleGraph(total_household_pop)
vertex_pointer = 1
for household in households_composition
num_vertices = sum(household)
- for v in vertex_pointer:(vertex_pointer + num_vertices - 1)
- for w in vertex_pointer:(vertex_pointer + num_vertices - 1)
- if v != w
- add_edge!(network,v,w)
- end
+ household_pointer = 0
+ for (k,size) in enumerate(household)
+ @inbounds population_list[vertex_pointer+household_pointer:vertex_pointer+household_pointer+size-1] .= AgentDemographic(k)
+ household_pointer += size
+ end
+ for v in vertex_pointer:(vertex_pointer + num_vertices - 1), w in vertex_pointer:(vertex_pointer + num_vertices - 1)
+ if v != w
+ add_edge!(network,v,w)
end
end
vertex_pointer+=num_vertices
end
- return network
+ return network,population_list
end
@@ -59,19 +65,8 @@ index_vectors: Vector of 3 vectors, each of which contains the indexes of all th
"""
function generate_population(num_households)
households_composition = sample_household_data(num_households)
-
- household_networks = complete_graph_from_households_composition(households_composition)
-
- #test the complete_graph_from_households_composition function against the mapreduce version to ensure it is correct.
- @c_assert all(adjacency_matrix(household_networks) .== adjacency_matrix( mapreduce(LightGraphs.complete_graph,blockdiag, sum.(households_composition); init = SimpleGraph())))
-
- households = map( l -> [fill(AgentDemographic(i),n) for (i,n) in enumerate(l)],households_composition) |>
- l -> reduce(vcat,l) |>
- l -> reduce(vcat,l)
- population_list = reduce(vcat,households)
+ household_networks,population_list = complete_graph_from_households_composition(households_composition)
index_vectors = [findall(x -> x == AgentDemographic(i), population_list) for i in 1:(AgentDemographic.size-1)]
-
-
return (;
population_list,
household_networks,
@@ -81,7 +76,7 @@ end
"""
-Defines pretty printing methods so that `display(AgentStatus)` is more readable.
+Defines pretty printing methods so that `display(s::AgentStatus)` is more readable.
"""
function Base.show(io::IO, status::AgentStatus)
if status == Susceptible
@@ -96,7 +91,7 @@ function Base.show(io::IO, status::AgentStatus)
end
"""
-Defines pretty printing methods so that `display(AgentDemographic)` is more readable.
+Defines pretty printing methods so that `display(s::AgentDemographic)` is more readable.
"""
function Base.show(io::IO, status::AgentDemographic)
if status == Young
diff --git a/CovidAlertVaccinationModel/src/ABM/solve.jl b/CovidAlertVaccinationModel/src/ABM/solve.jl
index 4470117..737121f 100644
--- a/CovidAlertVaccinationModel/src/ABM/solve.jl
+++ b/CovidAlertVaccinationModel/src/ABM/solve.jl
@@ -1,10 +1,19 @@
-# using LoopVectorization
+
+
+
+
function contact_weight(p, contact_time)
return 1 - (1-p)^contact_time
end
+
+
+
Base.@propagate_inbounds @views function update_alert_durations!(t,modelsol)
+#remove Base.@propagate_inbounds if you get segfaults
@unpack notification_parameter = modelsol.params
@unpack time_of_last_alert, app_user_index,inf_network,covid_alert_times,app_user = modelsol
+
+
for (i,node) in enumerate(modelsol.app_user_index), mixing_graph in modelsol.inf_network.graph_list[t]
for j in 2:14
covid_alert_times[j-1,i] = covid_alert_times[j,i] #shift them all back
@@ -15,13 +24,17 @@ Base.@propagate_inbounds @views function update_alert_durations!(t,modelsol)
covid_alert_times[end,i] += get_weight(mixing_graph,GraphEdge(node,j)) #add the contact times for today to the back
end
end
- if rand(RNG) < 1 - (1- notification_parameter)^sum(covid_alert_times[:,i])
+
+ covid_alert_total_exposures = 1 - (1 - notification_parameter) ^ sum(covid_alert_times[:,i])
+ if rand(RNG) < covid_alert_total_exposures
time_of_last_alert[i] = t
end
end
end
Base.@propagate_inbounds @views function update_infection_state!(t,modelsol)
+#remove Base.@propagate_inbounds if you get segfaults
+
@unpack base_transmission_probability,immunization_loss_prob,recovery_rate = modelsol.params
@unpack u_inf,u_vac,u_next_inf,u_next_vac,demographics,inf_network,status_totals = modelsol
@@ -65,6 +78,7 @@ Base.@propagate_inbounds @views function update_infection_state!(t,modelsol)
# display(u_next_inf)
end
Base.@propagate_inbounds @views function update_vaccination_opinion_state!(t,modelsol,total_infections)
+#remove Base.@propagate_inbounds if you get segfaults
@unpack Ï€_base, η,γ, κ, ω, Ï, ω_en,Ï_en,γ,β = modelsol.params
@unpack demographics,time_of_last_alert, nodes, soc_network,u_vac,u_next_vac,app_user,app_user_list = modelsol
diff --git a/CovidAlertVaccinationModel/src/data.jl b/CovidAlertVaccinationModel/src/data.jl
index a84afeb..5532743 100644
--- a/CovidAlertVaccinationModel/src/data.jl
+++ b/CovidAlertVaccinationModel/src/data.jl
@@ -35,20 +35,20 @@ function find_household_composition(df_row)
age_distribution[age_resp_to_bin[df_row[:AGERESP]]] += 1
return SVector{3}(age_distribution)
end
-function sample_household_data(n)
+function read_household_data()
f = readdlm(joinpath(PACKAGE_FOLDER,"data/csv/home_compositions.csv"), ',')
df = DataFrame([f[1,i] => f[2:end, i] for i = 1:length(f[1,:])])
weight_vector::Vector{Float64} = df[!,:WGHT_PER]/sum(df[!,:WGHT_PER])
households = map(find_household_composition,eachrow(df))
- return sample(RNG,households, Weights(weight_vector),n)
- # https://www.publichealthontario.ca/-/media/documents/ncov/epi/covid-19-severe-outcomes-ontario-epi-summary.pdf?la=en
+ return (;households,weight_vector)
+end
+
+function sample_household_data(n)
+ return sample(RNG,household_data.households, Weights(household_data.weight_vector),n)
end
+
function get_household_data_proportions()
- f = readdlm(joinpath(PACKAGE_FOLDER,"data/csv/home_compositions.csv"), ',')
- df = DataFrame([f[1,i] => f[2:end, i] for i = 1:length(f[1,:])])
- weight_vector::Vector{Float64} = df[!,:WGHT_PER]/sum(df[!,:WGHT_PER])
- households = map(find_household_composition,eachrow(df))
- households_by_demographic_sum = sum.([map(l-> l[i], households) for i in 1:3])
+ households_by_demographic_sum = sum.([map(l-> l[i],household_data.households) for i in 1:3])
return households_by_demographic_sum./sum(households_by_demographic_sum)
# https://www.publichealthontario.ca/-/media/documents/ncov/epi/covid-19-severe-outcomes-ontario-epi-summary.pdf?la=en
end
--
GitLab