Mitsubishi Report LP versus IP experiment

gistfile1.txt

using Base.Test, SchedulingModels, JuMP, Gurobi

np = SchedulingModels.readnetwork("config/multiline-arrivals3.json")

# Integer Optimization based on Arrival Model
model = arrivalmodel(np, ScheduleOptions(), solver=Gurobi.GurobiSolver(TimeLimit=1000))
@time status = JuMP.solve(model.model)

#   ...
#   1368  1076 151569.952   38  315 199238.000 151569.952  23.9%  97.3  984s
#   1370  1077 181092.352  210  319 199238.000 151578.308  23.9%  97.1  985s
# H 1374  1016                    199193.00000 151595.802  23.9%  96.8  994s
#   1377  1018 151600.349   85  316 199193.000 151600.349  23.9%  96.6  995s

# Cutting planes:
#   Gomory: 7
#   MIR: 1868
#   Flow cover: 1881

# Explored 1382 nodes (374250 simplex iterations) in 1000.03 seconds
# Thread count was 4 (of 4 available processors)

# Solution count 10: 199193 199238 199643 ... 212865

# Time limit reached
# Best objective 1.991930000000e+05, best bound 1.516027781081e+05, gap 23.8915%
# WARNING: Not solved to optimality, status: UserLimit
# 1000.125847 seconds (220 allocations: 32.567 MiB, 0.01% gc time)
# :UserLimit

scheduleIP = SchedulingModels.schedule(model)

# julia> sum(scheduleIP, 2)
# 6×1 Array{Int64,2}:
#  4
#  4
#  7
#  2
#  6
#  7

# Linear Optimization based on Network Model
lpmodel = networkmodel(np, ScheduleOptions(kind=:rate), solver=Gurobi.GurobiSolver())
@time lpstatus = JuMP.solve(lpmodel.model)

# Solved with barrier
# Solved in 106029 iterations and 284.07 seconds
# Optimal objective  3.649595545e+04
# 290.170663 seconds (171 allocations: 1.048 GiB, 1.23% gc time)
# :Optimal

scheduleLP = SchedulingModels.schedule(lpmodel)
# julia> round(Int, sum(scheduleLP,2))
# 6×1 Array{Int64,2}:
#  5
#  3
#  8
#  2
#  6
#  6