diff --git a/src/jl/MCP_heuristics.jl b/src/jl/MCP_heuristics.jl
index 6e01c8fbf450f040ff275ff297fbe69fde93cfe4..035f66c498b651680ec449dbfdf269dafc0027f5 100644
--- a/src/jl/MCP_heuristics.jl
+++ b/src/jl/MCP_heuristics.jl
@@ -334,14 +334,14 @@ end
function threshold_greedy_algorithm(D::Array{Float64,2}, c::Float64, n::Float64)
W, L = size(D)
- x_incumbent = zeros(Float64, L)
+ n = convert(Int64, n)
ind_compl_incumbent = [i for i in 1:L]
- ind_incumbent = []
+ ind_incumbent = Vector{Int64}(undef, n)
Dx_incumbent = zeros(Float64, W)
- y_incumbent = zeros(Float64, W)
obj_incumbent = 0
Dx_tmp = Vector{Float64}(undef, W)
y_tmp = Vector{Float64}(undef, W)
+ ind_candidate_list = zeros(Int64, L)
locations_added, threshold = 0, 0
@inbounds while locations_added < n
if locations_added < c
@@ -350,28 +350,29 @@ function threshold_greedy_algorithm(D::Array{Float64,2}, c::Float64, n::Float64)
else
obj_candidate = obj_incumbent
end
- ind_candidate_list = Vector{Int64}(undef, 0)
- @inbounds for ind in setdiff(ind_compl_incumbent, ind_incumbent)
+ ind_candidate_pointer = 1
+ @inbounds for ind in ind_compl_incumbent
Dx_tmp .= Dx_incumbent .+ view(D, :, ind)
y_tmp .= Dx_tmp .>= threshold
obj_tmp = sum(y_tmp)
if obj_tmp > obj_candidate
- ind_candidate_list = [ind]
+ ind_candidate_pointer = 1
+ ind_candidate_list[ind_candidate_pointer] = ind
obj_candidate = obj_tmp
+ ind_candidate_pointer += 1
elseif obj_tmp == obj_candidate
- ind_candidate_list = union(ind, ind_candidate_list)
+ ind_candidate_list[ind_candidate_pointer] = ind
+ ind_candidate_pointer += 1
end
end
- ind_candidate = sample(ind_candidate_list)
- ind_incumbent = union(ind_incumbent, ind_candidate)
+ ind_candidate = sample(view(ind_candidate_list, 1:ind_candidate_pointer-1))
+ ind_incumbent[locations_added+1] = ind_candidate
+ filter!(a -> a != ind_candidate, ind_compl_incumbent)
Dx_incumbent .= Dx_incumbent .+ view(D, :, ind_candidate)
- y_incumbent .= Dx_incumbent .>= c
- obj_incumbent = sum(y_incumbent)
+ obj_incumbent = obj_candidate
locations_added += 1
end
- x_incumbent[ind_incumbent] .= 1.
- return x_incumbent, obj_incumbent
-
+ return ind_incumbent, obj_incumbent
end
function time_threshold_greedy_algorithm(D::Array{Float64,2}, c::Float64, n::Float64)
@@ -397,16 +398,15 @@ end
function stochastic_threshold_greedy_algorithm(D::Array{Float64,2}, c::Float64, n::Float64, p::Float64)
W, L = size(D)
- s = convert(Int64, round((L/n)*p))
- x_incumbent = zeros(Float64, L)
+ s = convert(Int64, round(L*p))
random_ind_set = Vector{Int64}(undef, s)
ind_compl_incumbent = [i for i in 1:L]
- ind_incumbent = []
+ ind_incumbent = Vector{Int64}(undef, convert(Int64, n))
Dx_incumbent = zeros(Float64, W)
- y_incumbent = zeros(Float64, W)
obj_incumbent = 0
Dx_tmp = Vector{Float64}(undef, W)
y_tmp = Vector{Float64}(undef, W)
+ ind_candidate_list = zeros(Int64, L)
locations_added, threshold = 0, 0
@inbounds while locations_added < n
if locations_added < c
@@ -415,30 +415,30 @@ function stochastic_threshold_greedy_algorithm(D::Array{Float64,2}, c::Float64,
else
obj_candidate = obj_incumbent
end
- ind_candidate_list = Vector{Int64}(undef, 0)
- random_ind_set .= sample(ind_compl_incumbent, s, replace=false)
+ ind_candidate_pointer = 1
+ sample!(ind_compl_incumbent, random_ind_set, replace=false)
@inbounds for ind in random_ind_set
Dx_tmp .= Dx_incumbent .+ view(D, :, ind)
y_tmp .= Dx_tmp .>= threshold
obj_tmp = sum(y_tmp)
if obj_tmp > obj_candidate
- ind_candidate_list = [ind]
+ ind_candidate_pointer = 1
+ ind_candidate_list[ind_candidate_pointer] = ind
obj_candidate = obj_tmp
+ ind_candidate_pointer += 1
elseif obj_tmp == obj_candidate
- ind_candidate_list = union(ind, ind_candidate_list)
+ ind_candidate_list[ind_candidate_pointer] = ind
+ ind_candidate_pointer += 1
end
end
- ind_candidate = sample(ind_candidate_list)
- ind_incumbent = union(ind_incumbent, ind_candidate)
- ind_compl_incumbent = setdiff(ind_compl_incumbent, ind_candidate)
+ ind_candidate = sample(view(ind_candidate_list, 1:ind_candidate_pointer-1))
+ ind_incumbent[locations_added+1] = ind_candidate
+ filter!(a -> a != ind_candidate, ind_compl_incumbent)
Dx_incumbent .= Dx_incumbent .+ view(D, :, ind_candidate)
- y_incumbent .= Dx_incumbent .>= c
- obj_incumbent = sum(y_incumbent)
+ obj_incumbent = obj_candidate
locations_added += 1
end
- x_incumbent[ind_incumbent] .= 1.
- return x_incumbent, obj_incumbent
-
+ return ind_incumbent, obj_incumbent
end
function stochastic_threshold_greedy_algorithm_trajectory(D::Array{Float64,2}, c::Float64, n::Float64, p::Float64)
diff --git a/src/jl/SitingHeuristics.jl b/src/jl/SitingHeuristics.jl
index 7790a320d8fe9a776495a64411f36e82e723ae77..6b9c88cee954dd2b90ba2fe85ab12cd2b99485a5 100644
--- a/src/jl/SitingHeuristics.jl
+++ b/src/jl/SitingHeuristics.jl
@@ -45,7 +45,7 @@ function main_MIRSA(index_dict, deployment_dict, D, c, N, I, E, T_init, R, run,
if run == "MIR"
x_sol, LB_sol, obj_sol = Array{Float64, 2}(undef, R, L), Array{Float64, 1}(undef, R), Array{Float64, 2}(undef, R, I)
- x_init = solve_MILP_partitioning(D, c, n_partitions, index_dict, "Gurobi")
+ x_init = solve_MILP(D, c, n, "Gurobi")
for r = 1:R
x_sol[r, :], LB_sol[r], obj_sol[r, :] = simulated_annealing_local_search(D, c, n, N, I, E, x_init, T_init, legacy_index)
@@ -56,7 +56,7 @@ function main_MIRSA(index_dict, deployment_dict, D, c, N, I, E, T_init, R, run,
x_sol, LB_sol, obj_sol = Array{Float64, 2}(undef, R, L), Array{Float64, 1}(undef, R), Array{Float64, 2}(undef, R, I)
x_init = zeros(Int64, L)
ind_set = [l for l in 1:L]
- n_while = n_partitions[1]
+ n_while = n
while n_while > 0
loc = sample(ind_set)
@@ -153,38 +153,3 @@ function main_RAND(deployment_dict, D, c, I, R, run)
return x_sol, LB_sol
end
-
-# Local Search algorithm
-function main_LSEA(index_dict, deployment_dict, D, c, N, I, E, run)
-
- index_dict = Dict([(convert(Int64, k), convert(Int64, index_dict[k])) for k in keys(index_dict)])
- deployment_dict = Dict([(convert(Int64, k), convert(Int64, deployment_dict[k])) for k in keys(deployment_dict)])
- D = convert.(Float64, D)
-
- c = convert(Float64, c)
- N = convert(Int64, N)
- I = convert(Int64, I)
- E = convert(Int64, E)
- T_init = convert(Float64, T_init)
- R = convert(Int64, R)
- run = string(run)
-
- W, L = size(D)
-
- P = maximum(values(index_dict))
- n_partitions = [deployment_dict[i] for i in 1:P]
-
- if run == "GLS"
- x_sol, LB_sol, obj_sol = Array{Float64, 2}(undef, R, L), Array{Float64, 1}(undef, R), Array{Float64, 2}(undef, R, I)
- x_init = solve_MILP_partitioning(D, c, n_partitions, index_dict, "Gurobi")
- for r = 1:R
- x_sol[r, :], LB_sol[r], obj_sol[r, :] = greedy_local_search_partition(D, c, n_partitions, N, I, E, x_init, index_dict)
- end
- else
- println("No such run available.")
- throw(ArgumentError)
- end
-
- return x_sol, LB_sol, obj_sol
-
-end