forked from CMA-ES/c-cmaes
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcmaes_initials.par
105 lines (85 loc) · 5 KB
/
cmaes_initials.par
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
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
function number 1 2 5 6 8 23 # the first number is read by example_restarts.c, see there.
restarts 0 # read by example_restarts.c, remark that restarts
# with the same initialX might be sub-optimal
#
# Input parameter file for cmaes_t.
# Comments start with '#' or '%' until end of line.
# Actual recent parameter setting is written to file actparcmaes.par.
#
## --- OBLIGATORY SETTINGS
# these settings, if read, overwrite the input values to cmaes_init
N 22 # Problem dimension, overwrites parameter in cmaes_init
initialX 1: # Initial search point. Syntax: 1==read one number, see also end of file.
0.5e0 # these are the read value(s)!
typicalX 1: # Typical search point (useful for restarts, cave: boundaries), overwritten by initialX
0.0 # these are the read value(s)
initialStandardDeviations 1: # 1==read only one number for all coordinates
0.3 # numbers should not differ by orders of magnitude
# should be roughly 1/4 of the search interval
# this number essentially influences the global
# search ability (ie. the horizon where to search
# at all) on multimodal functions
## --- OPTIONAL SETTINGS
# stop-Parameters can be changed online via cmaes_signals.par
stopMaxFunEvals 1e299 # max number of f-evaluations, 900*(N+3)*(N+3) is default
# fac*maxFunEvals 1e0 # multiplies stopMaxFunEvals by read number, default is 1
stopMaxIter 1e299 # max number of iterations (generations), inf is default
# stopFitness 1e-9 # stop if function value is smaller than stopFitness
# commented == never stop on function value (default)
stopTolFun 1e-12 # stop if function value differences are
# smaller than stopTolFun, default=1e-12
stopTolFunHist 1e-13 # stop if function value differences of best values are
# smaller than stopTolFunHist, default was 0
stopTolX 1e-11 # stop if step sizes/steps in x-space are
# smaller than TolX, default=0
stopTolUpXFactor 1e3 # stop if std dev increases more than by TolUpXFactor, default 1e3
seed 0 # 0 == by time, also regard maxTimeFractionForEigendecomposition
# below, as for values smaller than one the outcome might not be
# exactly reproducible even with the same seed
# diffMinChange 1 # Minimal coordinate wise standard deviation. Syntax see below.
# 1e-299 # Interferes with stopTolX termination criterion! default=0
## --- internal CPU-time related settings
maxTimeFractionForEigendecompostion 1 # maximal CPU-time fraction for eigensystem
# decomposition. Large values (up to one) are better
# w.r.t. the number of function evaluations to reach a
# certain function value. Only >=1 yields exactly
# reproducible results.
# 0.2==20% should often be faster than larger values
# w.r.t. the CPU-time to reach a certain function value.
# updatecov 1 # default is updating the eigensystem after
# every 1/ccov/N/10-th generation.
# fac*updatecov 3 # multiplier for updatecov
# resume allcmaes.dat # reads restart distribution from given file
## --- Strategy internal parameter ---
## default values are set in readpara_SupplementDefaults()
## -- Selection related parameters
# lambda 100 # number of offspring == samplesize
# mu
# weights log # possible values: log (==default),
# lin (==linearely decreasing), or equal
## -- Adaptation (distribution estimation) related parameters
# diagonalCovarianceMatrix -1 # number of iterations, 1<==>always, -1<==>2+100*N/sqrt(lambda)
# fac*damp 1 # increase or decrease damping for step size control.
# ccumcov 1 # default is 4/(N+4), 1 means no cumulation for p_c.
# mucov 1 # 1 means only rank 1 update of C
# fac*ccov 1 # multiplier for default learning rate for cov. matrix
## --- Syntax for input vectors xstart, sigma, mincoorstddev
#
# N 7 # dimension must be defined before
# xstart 3 : # read 3 numbers from next lines, the colon is superfluous
# 0.2 0.3
# 0.4 0.5 0.6
# 0.7
#
# reads the first three numbers starting from the
# line following the keyword "xstart" and recycles
# these numbers (or cut them) to length N, resulting in
# xstart=[0.2 0.3 0.4 0.2 0.3 0.4 0.2]
# No comments are allowed between the numbers. I.e.
#
# xstart 3 22 anything here is ok, 22 is disregarded # still ok
# 0.2 0.3 # this comment fails, if >2 numbers to be read
# 0.4 0.5 0.6
#
# would fail due to the comment between second and third number to
# be read.