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
87
88
89
90
|
/**
* @file sa_test.cpp
* @author Zhihao Lou
* @author Marcus Edel
* @author Conrad Sanderson
*
* ensmallen is free software; you may redistribute it and/or modify it under
* the terms of the 3-clause BSD license. You should have received a copy of
* the 3-clause BSD license along with ensmallen. If not, see
* http://www.opensource.org/licenses/BSD-3-Clause for more information.
*/
#if defined(ENS_USE_COOT)
#include <armadillo>
#include <bandicoot>
#endif
#include <ensmallen.hpp>
#include "catch.hpp"
#include "test_function_tools.hpp"
using namespace ens;
using namespace ens::test;
// The Generalized-Rosenbrock function is a simple function to optimize.
TEMPLATE_TEST_CASE("SA_GeneralizedRosenbrockFunction", "[SA]",
ENS_ALL_CPU_TEST_TYPES)
{
typedef typename TestType::elem_type ElemType;
size_t dim = 10;
GeneralizedRosenbrockFunction f(dim);
size_t iteration = 0;
ElemType result = DBL_MAX;
TestType coordinates;
while (result > 10 * Tolerances<TestType>::Obj)
{
ExponentialSchedule schedule;
// The convergence is very sensitive to the choices of maxMove and initMove.
SA<ExponentialSchedule> sa(schedule, 1000000, 1000., 1000, 100,
Tolerances<TestType>::Obj / 1000, 3, 1.5, 0.5, 0.3);
coordinates = f.template GetInitialPoint<TestType>();
result = sa.Optimize(f, coordinates);
// No more than three tries, or five for low-precision.
const size_t limit = (sizeof(ElemType) < 4) ? 5 : 3;
REQUIRE(iteration <= limit);
++iteration;
}
// Use type-specific tolerances.
REQUIRE(result == Approx(0.0).margin(10 * Tolerances<TestType>::LargeObj));
for (size_t j = 0; j < dim; ++j)
{
REQUIRE(coordinates(j) ==
Approx(1.0).epsilon(Tolerances<TestType>::LargeCoord));
}
}
// The Rosenbrock function is a simple function to optimize.
TEMPLATE_TEST_CASE("SA_RosenbrockFunction", "[SA]", ENS_ALL_TEST_TYPES)
{
ExponentialSchedule schedule;
// The convergence is very sensitive to the choices of maxMove and initMove.
SA<> sa(schedule, 1000000, 1000., 1000, 100, Tolerances<TestType>::Obj, 3,
1.5, 0.3, 0.3);
// Large tolerances are necessary due to the sensitivity of convergence.
FunctionTest<RosenbrockFunction, TestType>(sa,
10 * Tolerances<TestType>::LargeObj,
10 * Tolerances<TestType>::LargeCoord,
5);
}
/**
* The Rastrigrin function, a (not very) simple nonconvex function. It has very
* many local minima, so finding the true global minimum is difficult.
*/
TEMPLATE_TEST_CASE("SA_RastrigrinFunction", "[SA]", ENS_ALL_TEST_TYPES)
{
// Simulated annealing isn't guaranteed to converge (except in very specific
// situations). If this works 1 of 4 times, I'm fine with that. All I want
// to know is that this implementation will escape from local minima.
ExponentialSchedule schedule;
// The convergence is very sensitive to the choices of maxMove and initMove.
// SA<> sa(schedule, 2000000, 100, 50, 1000, 1e-12, 2, 2.0, 0.5, 0.1);
SA<> sa(schedule, 2000000, 100, 50, 1000, Tolerances<TestType>::Obj, 2, 2.0,
0.5, 0.1);
FunctionTest<RastriginFunction, TestType>(sa, Tolerances<TestType>::LargeObj,
Tolerances<TestType>::LargeCoord, 4);
}
|
