Skip to content

Commit

Permalink
modified the decision functions for branching variable selection
Browse files Browse the repository at this point in the history
  • Loading branch information
@LeChStan
LeChStan committed Aug 4, 2024
1 parent f422ecc commit 4351da9
Showing 1 changed file with 31 additions and 26 deletions.
57 changes: 31 additions & 26 deletions src/pseudo_branching.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -37,44 +37,36 @@ function Bonobo.get_branching_variable(
pseudos::SparseMatrixCSC{Float64, Int64},
branch_tracker::SparseMatrixCSC{Int64, Int64},
) where BLMO <: BoundedLinearMinimizationOracle
#the following should create the arrays and vectors only on first function call and then use existing one
# local pseudos = sparse(
# repeat(Boscia.get_integer_variables(branching.bounded_lmo), 2),
# vcat(ones(length(Boscia.get_integer_variables(branching.bounded_lmo))), 2*ones(length(Boscia.get_integer_variables(branching.bounded_lmo)))),
# ones(2 * length(Boscia.get_integer_variables(branching.bounded_lmo)))
# )
# local branch_tracker = sparse(
# repeat(Boscia.get_integer_variables(branching.bounded_lmo), 2),
# vcat(ones(length(Boscia.get_integer_variables(branching.bounded_lmo))), 2*ones(length(Boscia.get_integer_variables(branching.bounded_lmo)))),
# ones(Int64, 2 * length(Boscia.get_integer_variables(branching.bounded_lmo)))
# )

#local call_tracker = 0
strategy_switch = branching.iterations_until_stable + 1
best_idx = -1
all_stable = true
branching_candidates = Int[]# this shall contain the indices of the potential branching variables
values = Bonobo.get_relaxed_values(tree, node)
branch_counter = []
for idx in Bonobo.get_branching_indices(tree.root)
# branch_counter = []
# for idx in Bonobo.get_branching_indices(tree.root)
for idx in tree.branching_indices
value = values[idx]
if !Bonobo.is_approx_feasible(tree, value)
push!(branching_candidates, idx)
push!(branch_counter, (branch_tracker[idx, 1], branch_tracker[idx, 2]))
# push!(branch_counter, (branch_tracker[idx, 1], branch_tracker[idx, 2]))
if branch_tracker[idx, 1] < strategy_switch || branch_tracker[idx, 2] < strategy_switch# check if pseudocost is stable for this idx
all_stable = false
end
end
end
if rand() > 0.99# for debugging on if strategy behaves as intended
println("\n branch_counter")
println(branch_counter)
end
# if rand() > 0.5# for debugging on if strategy behaves as intended
# println("\n branch_counter")
# println(branch_counter)
# end
if node.parent_lower_bound_base != Inf# if this node is a result of branching on some variable then update pseudocost of corresponding branching variable
#println("if clause of update")
idx = node.branched_on
update = (tree.root.problem.f(values) - node.dual_gap) - node.parent_lower_bound_base
update = update / node.distance_to_int
# if rand() > 0.99
# update += 0.1
# end
if isinf(update)
@debug "update is $(Inf)"
end
Expand All @@ -95,7 +87,6 @@ function Bonobo.get_branching_variable(
# println(sum(branch_tracker))
# end
end

length(branching_candidates) == 0 && return best_idx
length(branching_candidates) == 1 && return branching_candidates[1]
if !all_stable# THEN Use Most Infeasible
Expand All @@ -108,7 +99,6 @@ function Bonobo.get_branching_variable(
max_distance_to_feasible = distance_to_feasible
end
end
#println(best_idx)
return best_idx
else
# Pseudocosts have stabilized
Expand All @@ -118,18 +108,33 @@ function Bonobo.get_branching_variable(

if branching.decision_function == "weighted_sum"
#println(branching.decision_function)
branching_scores = map(idx-> ((1 - branching.μ) * min(pseudos[idx, 1] * (values[idx] - floor(values[idx])), pseudos[idx, 2] * (ceil(values[idx]) - values[idx])) + branching.μ * max(pseudos[idx, 1] * (values[idx] - floor(values[idx])), pseudos[idx, 2] * (ceil(values[idx]) - values[idx]))),
branching_scores = map(idx-> ((1 - branching.μ) * min((pseudos[idx, 1] - 1) * (values[idx] - floor(values[idx])), (pseudos[idx, 2] - 1) * (ceil(values[idx]) - values[idx])) + branching.μ * max((pseudos[idx, 1] - 1) * (values[idx] - floor(values[idx])), (pseudos[idx, 2] - 1) * (ceil(values[idx]) - values[idx]))),
branching_candidates)

elseif branching.decision_function == "product"
#println(branching.decision_function)
branching_scores = map(idx-> max(pseudos[idx, 1] * (values[idx] - floor(values[idx])), branching.μ) * max(pseudos[idx, 2] * (ceil(values[idx]) - values[idx]), branching.μ),
branching_scores = map(idx-> max((pseudos[idx, 1] - 1) * (values[idx] - floor(values[idx])), branching.μ) * max((pseudos[idx, 2] - 1) * (ceil(values[idx]) - values[idx]), branching.μ),
branching_candidates)

end
branching_scores = sparsevec(branching_candidates, branching_scores)
#display(branching_scores)
best_idx = argmax(branching_scores)
best_idx = argmax(branching_scores)
# max_distance_to_feasible = 0.0
# best_idx_base = -1
# for i in branching_candidates
# value = values[i]
# distance_to_feasible = Bonobo.get_distance_to_feasible(tree, value)
# if distance_to_feasible > max_distance_to_feasible
# best_idx_base = i
# max_distance_to_feasible = distance_to_feasible
# end
# end
# if best_idx != best_idx_base
# println(best_idx, best_idx_base)
# println("different branching decision made")
# end

#println(pseudos[best_idx, 1], pseudos[best_idx,2])
#println(best_idx)
return best_idx
end
Expand All @@ -142,7 +147,7 @@ function update_avg(new_val::Float64, avg::Float64, N::Int)
if N > 1
return 1/(N+1) * (N * (avg - 1) + new_val) + 1 #
else
return new_val+ avg # note that we initialize the pseudo costs with 1 as such we need to shift pseudocosts correspondingly
return new_val + avg # note that we initialize the pseudo costs with 1 as such we need to shift pseudocosts correspondingly
end
end
#pseudos = Dict{Int,Array{Float64}}(i=>zeros(2) for idx in get_integer_variables(blmo))
Expand Down

0 comments on commit 4351da9

Please sign in to comment.