balans/main_balans.py at main · coin-or/balans · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
# ALNS for adaptive large neighborhood
from alns.accept import SimulatedAnnealing
from alns.select import MABSelector
from alns.stop import MaxIterations, MaxRuntime

# MABWiser for contextual multi-armed bandits
from mabwiser.mab import LearningPolicy

# Balans meta-solver for solving mixed integer programming problems
from balans.solver import Balans, DestroyOperators, RepairOperators

import numpy as np

# Destroy operators
destroy_ops = [DestroyOperators.Crossover,
               # DestroyOperators.Dins,
               DestroyOperators.Mutation_25,
               DestroyOperators.Mutation_50,
               DestroyOperators.Mutation_75,
               DestroyOperators.Local_Branching_10,
               DestroyOperators.Local_Branching_25,
               DestroyOperators.Local_Branching_50,
               DestroyOperators.Proximity_005,
               DestroyOperators.Proximity_015,
               DestroyOperators.Proximity_030,
               # DestroyOperators.Random_Objective
               DestroyOperators.Rens_25,
               DestroyOperators.Rens_50,
               DestroyOperators.Rens_75,
               DestroyOperators.Rins_25,
               DestroyOperators.Rins_50,
               DestroyOperators.Rins_75]

# Repair operators
repair_ops = [RepairOperators.Repair]

# Rewards
reward_best, reward_better, reward_accept, reward_reject = 1, 1, 0, 0

# Bandit selector
selector = MABSelector(scores=[reward_best, reward_better, reward_accept, reward_reject],
                       num_destroy=len(destroy_ops),
                       num_repair=len(repair_ops),
                       learning_policy=LearningPolicy.ThompsonSampling())

# Acceptance criterion
# accept = HillClimbing()
accept = SimulatedAnnealing(start_temperature=20, end_temperature=1, step=0.1)

# Stopping condition
# stop = MaxRuntime(300)
stop = MaxIterations(10)

# Balans
# To use the default config shipped with the package:
#   from balans.utils import ConfigFactory
#   balans = Balans(config=ConfigFactory.DEFAULT_CONFIG_PATH)
balans = Balans(destroy_ops=destroy_ops,
                repair_ops=repair_ops,
                selector=selector,
                accept=accept,
                stop=stop,
                seed=1283,
                n_mip_jobs=4,
                mip_solver="scip",
                timelimit_first_solution=10,
                timelimit_alns_iteration=5,
                timelimit_local_branching_iteration=10,
                timelimit_crossover_random_feasible=5,
                big_m=1000)

# Run
instance_path = "tests/data/correctness_mk_5_maximize.lp"
result = balans.solve(instance_path)

# print("Best solution:", result.best_state.solution())
print("Best objective:", result.best_state.objective())
print("Objective trace:", result.statistics.objectives)
# print("Runtime trace:", result.statistics.runtimes)
print("Runtime trace (cumulative):", np.cumsum(result.statistics.runtimes))
print("Operator to reward (best, better, accept, reject) counts:", dict(result.statistics.destroy_operator_counts))

# The installed CLI entry point lives in balans/main.py.
# from balans.main_balans import main
# if __name__ == "__main__":
#     main()