Compare ARCOPT with results presented in Antonino’s thesis
This directory contains work to compare arcopt with results preseneted in
Antonino Del Gatto’s (ADG) thesis. The process went like this:
Search for problems in the CUTEr database with names that match those in
ADG’s thesis. ADG’s test set included 37 problems. I found matches in the
cuter set for 30 problems. These lists can be seen in the notes section that
follows.
I ran a set of tests to evaluate the quality and functionality of the test
problems. I first check that the problem doesn’t just crash. Second, I test
the derivative with finite differences. Most of the problems passed most of
the tests. The problem named WATSON did not appear to do so well with a
relative error in the hessian check of 1e-3.
I then attempted to solve the 30 matching problems with ARCOPT. The results
effectively matched for 28 of the 30 problems. ARCOPT failed on 2 of the
problems, with one of those being WATSON. The comparison table can be seen
in the notes section of this file.
short description of files
file
note
doc
contains table from ADG’s thesis
runs
output from arcopt
adg_run1.m
run single problem with arcopt
adg_run_all.m
run all adg problem with arcopt
adg_setup.m
set up matlab environment
adg_test_problems.m
run tests on adg problems
solve_pdat.m
wrapper to cuter for 1 problem
solve_pdat_set.m
wrapper to cuter for many problems
adg_arcopt_compare.org
organized note with the final table
adg_results.org
arcopt results table
adg_set.org
ADG test set list
adg_test.org
results of tests on problems
notes
problems from thesis vs cuter set
adg p1
adg p2
size
iter
norm(g)
in cuter
param
ARGAUS
ARGAUS
3
3
2.47E-16
ARGAUSS
ARGQDN
ARGQDN50
5
2
0.00E+00
ARTRIG
ARTRIG10
10
11
7.45E-12
ARGTRIG
N=10
AVRIEL
AVRIEL3
2
2
4.44E-16
BARD70
BARD70
3
12
4.98E-16
BARD
BEAL58
BEAL58KO
2
11
2.90E-15
BEALE
BOOTH
BOOTH
2
2
3.55E-15
BOOTH
BOX66
BOX662HL
2
13
1.33E-16
BOX2
BRKMCC
BRKMCC
2
4
4.22E-15
BRKMCC
BROWND
BROWND
4
9
3.45E-11
BROWNDEN
BROY7D
BROY7D
60
28
2.22E-10
BROYDN7D
N/2=50
BRWNAL
BRWNL100
100
6
1.09E-11
BROWNAL
N=100
BRYBND
BRYBND18
100
8
1.39E-12
BRYBND
N=100
CLUSTR
CLUSTR
2
14
3.00E-18
CLUSTER
CRGLVY
CRGLY500
500
17
2.65E-14
CRAGGLVY
M=249
DIXON
DIXON
10
2
0.00E+00
DIXON3DQ
N=10
EXTRSN
EXTRA100
100
25
0.00E+00
EXTROSNB
N=100
FRDRTH
FRDRTHB3
50
7
1.75E-12
FREUROTH
N=50
GOTTFR
GOTTFR
2
20
1.17E-16
GOTTFR
HILBRT
HILBRT12
12
5
2.06E-23
HILBERTA
N=10
HIMM1
HIMM1
2
1
0.00E+00
HIMM25
HIMM25
2
1
0.00E+00
MANCIN
MANCIN50
50
4
5.36E-09
MANCINO
N=50
NONDIA
NONDIA20
20
26
8.66E-13
NONDIA
N=20
PENAL1
PEN1LN1
50
45
2.87E-17
PENALTY1
N=50
PENAL1
PEN1LN2
100
45
5.04E-16
PENALTY1
N=100
QUARTC
QUARTC
25
33
9.07E-13
QUARTC
N=25
SCHMVT
SCHMV500
500
4
4.51E-15
SCHMVETT
N=500
TDQUAD
TDQ500
500
2
9.86E-32
TOINT
PSPTOINT
50
12
5.61E-14
TOINTPSP
TRIDIA
TRLN100
100
2
6.86E-13
TRIDIA
N=100
WATSON
WATSON6
6
16
9.56E-12
WATSON
WOODS
WOODS
4
50
6.60E-14
WOODS
NS=1
WOODS
WOODS80
80
47
3.18E-13
WOODS
NS=25
XTX
XTX16
16
2
1.41E-31
XTX
XTX2
2
2
4.97E-32
ZANGWL
ZANGWL1
3
2
4.19E-29
ZANGWIL3
adg problems (removed problems missing in cuter db)