Passing options to WORHP
517 views
Skip to first unread message
Nicolò Robuschi
Mar 21, 2018, 6:25:08 PM3/21/18
to CasADi
Dear all,
I am currently using CasADi on Python 3.6 32bit, along with the SQP solver WORHP 1.10.
The installation of WORHP went smoothly, however I am experiencing some problem in passing the options to the solver.
For example, the syntax is this one:
opts_worhp = {}
opts_worhp["worhp.BFGSmethod"] = 1
opts_worhp["worhp.MaxIter"] = 5
opts_worhp["worhp.UserHM"] = False
opts_worhp["worhp.FidifHM"] = False
solver = nlpsol('solver', 'worhp', nlp_prob, opts_worhp)
but when the solver starts it gets stuck with default options. I am not able to set up a BFGS method nor to impose, for example, 5 iterations...it goes ahead to iterate with default options without showing any error.
I have also tried WORHP 1.11 and 1.07, but the problem persists in CasADi 3.3.0 and also in CasADi 3.4.0.
Do you have any idea?
Thanks a lot!
Best regards,
Nicolò
I am currently using CasADi on Python 3.6 32bit, along with the SQP solver WORHP 1.10.
The installation of WORHP went smoothly, however I am experiencing some problem in passing the options to the solver.
For example, the syntax is this one:
opts_worhp = {}
opts_worhp["worhp.BFGSmethod"] = 1
opts_worhp["worhp.MaxIter"] = 5
opts_worhp["worhp.UserHM"] = False
opts_worhp["worhp.FidifHM"] = False
solver = nlpsol('solver', 'worhp', nlp_prob, opts_worhp)
but when the solver starts it gets stuck with default options. I am not able to set up a BFGS method nor to impose, for example, 5 iterations...it goes ahead to iterate with default options without showing any error.
I have also tried WORHP 1.11 and 1.07, but the problem persists in CasADi 3.3.0 and also in CasADi 3.4.0.
Do you have any idea?
Thanks a lot!
Best regards,
Nicolò
Nicolò Robuschi
Mar 22, 2018, 6:20:06 AM3/22/18
to CasADi
Update: with the new release of CasADi 3.4.0 I am able to use WORHP also on Matlab 2016b. However, I am experiencing the same problem described in the previous post.
For example:
worhp_options = struct;
worhp_options.worhp.MaxIter = 5;
worhp_options.worhp.Timeout = 100;
nlp_solver = nlpsol('solver', 'worhp', nlp_prob, worhp_options);
the solver converge to an optimal solution in 52 iterations and Timeout time still the default one: 1800s.
Is it something related to WORHP itself? Or a compatibility issue between the solver and CasADi?
Thanks!
Best,
Nicolò
For example:
worhp_options = struct;
worhp_options.worhp.MaxIter = 5;
worhp_options.worhp.Timeout = 100;
nlp_solver = nlpsol('solver', 'worhp', nlp_prob, worhp_options);
the solver converge to an optimal solution in 52 iterations and Timeout time still the default one: 1800s.
Is it something related to WORHP itself? Or a compatibility issue between the solver and CasADi?
Thanks!
Best,
Nicolò
Joris Gillis
Apr 11, 2018, 3:48:09 AM4/11/18
to CasADi
Should be fixed now (3.4.1).
Nicolò Robuschi
Apr 11, 2018, 12:06:10 PM4/11/18
to CasADi
Dear Joris,
unfortunately I am facing the same type of error with WORHP. Maybe I am doing something wrong in declaring the options?
Thanks,
Nicolò
unfortunately I am facing the same type of error with WORHP. Maybe I am doing something wrong in declaring the options?
Thanks,
Nicolò
Joris Gillis
Apr 11, 2018, 5:18:57 PM4/11/18
to CasADi
Strange. Here's my test problem, and its output (for 3.4.1):
How does yours look?
Are you working on linux? It might be a different bug altogether then ( https://github.com/casadi/casadi/issues/2161 )
Best,
Joris
from casadi import *
x = MX.sym("x")
solver = nlpsol("solver","worhp",{"x":x,"g":x**2,"f":x**6},{"worhp": {"MaxIter":2}})
solver(lbg=1,ubg=2)
LICENSED FOR NON-PROFIT, ACADEMIC OR INSTRUCTIONAL USE.
YOU ARE NOT PERMITTED TO EXPLOIT THIS LICENSE COMMERCIALLY.
-------------------------------------------------------
This is WORHP 1.10-3, the European NLP-solver.
Use of WORHP is subject to terms and conditions.
Visit http://www.worhp.de for more information.
-------------------------------------------------------
Total number of variables ........................ 1
fixed variables 0
variables with lower bound only 0
variables with lower and upper bound 0
variables with upper bound only 0
Total number of box constraints .................. 0
Total number of other constraints ................ 2
equality constraints 0
inequality constraints with lower bound only 0
inequality constraints with lower and upper bound 1
inequality constraints with upper bound only 0
Gradient (user) 1/1 = 100.000%
Jacobian (user) 1/1 = 100.000%
Hessian (user) 1/1 = 100.000%
Algorithm Sequential Quadratic Programming
NLP MaxIter 2 Line Search Method Filter
QP MaxIter 500
LA solver MA97 (tol 1.0E-09, ref 10, ord METIS/AMD, scl none)
Tolerances:
Optimality (sKKT) 1.00E-06 (1.0E-03) IP ComTol 2.00E-07
Feasibility 1.00E-06 (1.0E-03) IP ResTol 4.00E-08
Complementarity 1.00E-06
Timeout 1800.000 seconds
ITER OBJ CON OPTI/COMPL FLAGS ALPHA |DX| REL PEN REG TIME
[ 0| 5] +0.00000000E+00 1.00000000E+00 0.00000000E+00 sc Uio 0.00E+00 0.00E+00 - - - 5.88E-04
[ 1| 1] +0.00000000E+00 1.00000000E+00 3.11747464E+07 so Uin 1.00E+00d 0.00E+00 1.0 - +0.6 1.42E-02
[ 2| 1] +0.00000000E+00 1.00000000E+00 3.03046132E+07 so Uin 1.00E+00d 0.00E+00 1.0 - +7.7 2.16E-02
Final values after iteration 2:
Final objective value ............. 0.0000000000E+00
Final constraint violation ........ 1.0000000000E+00
Final complementarity ............. 0.0000000000E+00 (0.0000000000E+00)
Final KKT conditions .............. 3.0304613191E+07 (0.0000000000E+00)
Unsuccessful termination: Maximum number of iterations reached.How does yours look?
Are you working on linux? It might be a different bug altogether then ( https://github.com/casadi/casadi/issues/2161 )
Best,
Joris
Nicolò Robuschi
Apr 12, 2018, 7:35:16 AM4/12/18
to CasADi
Dear Joris,
I am working on Windows 10 Pro, 64bit.
I have implemented your example on Python 3.6 32bit. This is the output of WORHP:
I am working on Windows 10 Pro, 64bit.
I have implemented your example on Python 3.6 32bit. This is the output of WORHP:
LICENSED FOR NON-PROFIT, ACADEMIC OR INSTRUCTIONAL USE.
YOU ARE NOT PERMITTED TO EXPLOIT THIS LICENSE COMMERCIALLY.
-------------------------------------------------------
This is WORHP 1.10-2160fae, the European NLP-solver.
Use of WORHP is subject to terms and conditions.
Visit http://www.worhp.de for more information.
-------------------------------------------------------
Total number of variables ........................ 1
fixed variables 0
variables with lower bound only 0
variables with lower and upper bound 0
variables with upper bound only 0
Total number of box constraints .................. 0
Total number of other constraints ................ 2
equality constraints 0
inequality constraints with lower bound only 0
inequality constraints with lower and upper bound 1
inequality constraints with upper bound only 0
Gradient (user) 1/1 = 100.000%
Jacobian (user) 1/1 = 100.000%
Hessian (user) 1/1 = 100.000%
Algorithm Sequential Quadratic Programming
NLP MaxIter 10000 Line Search Method Filter
QP MaxIter 500
LA solver MA97 (tol 1.0E-09, ref 10, ord METIS/AMD, scl none)
Tolerances:
Optimality (sKKT) 1.00E-06 (1.0E-03) IP ComTol 2.00E-07
Feasibility 1.00E-06 (1.0E-03) IP ResTol 4.00E-08
Complementarity 1.00E-06
Timeout 1800.000 seconds
ITER OBJ CON OPTI/COMPL FLAGS ALPHA |DX| REL PEN REG TIME
[ 0| 5] +0.00000000E+000 1.00000000E+000 0.00000000E+000 sc Uio 0.00E+000 0.00E+000 - - - 0.00E+000
[ 1| 1] +0.00000000E+000 1.00000000E+000 3.11747464E+007 so Uin 1.00E+000d 0.00E+000 1.0 - +0.6 6.30E-002
[ 2| 1] +0.00000000E+000 1.00000000E+000 3.03046132E+007 so Uin 1.00E+000d 0.00E+000 1.0 - +7.7 7.80E-002
[ 3| 1] +0.00000000E+000 1.00000000E+000 3.03046132E+007 so Uin 1.00E+000d 0.00E+000 1.0 - +7.6 9.40E-002
[ 4| 1] +0.00000000E+000 1.00000000E+000 3.03046132E+007 so Uin 1.00E+000d 0.00E+000 1.0 - +7.6 1.10E-001
Final values after iteration 4:
Final objective value ............. 0.0000000000E+000
Final constraint violation ........ 1.0000000000E+000
Final complementarity ............. 0.0000000000E+000 (0.0000000000E+000)
Final KKT conditions .............. 3.0304613191E+007 (0.0000000000E+000)
Unsuccessful termination: Minimum Stepsize reached in Armijo.
t_proc [s] t_wall [s] n_eval
nlp_f 0 0 2114
nlp_g 0 0 2120
nlp_grad_f 0 0 6
nlp_hess_l 0 0 5
nlp_jac_g 0 0 6
solver 0.156 0.156 1
As you can see MaxIter still the default one, but the same happens for all the other options that I am trying to tune.
The same happens in Matlab, which is 64bit; hence I would roule out a bitness issue.
Could be something related to the XML file in the lib folder? I tried to modify that file and also tried to delate it from the folder, but nothing changes.
Thank you!
Best,
Nicolò
[ 1| 1] +0.00000000E+000 1.00000000E+000 3.11747464E+007 so Uin 1.00E+000d 0.00E+000 1.0 - +0.6 6.30E-002
[ 2| 1] +0.00000000E+000 1.00000000E+000 3.03046132E+007 so Uin 1.00E+000d 0.00E+000 1.0 - +7.7 7.80E-002
[ 3| 1] +0.00000000E+000 1.00000000E+000 3.03046132E+007 so Uin 1.00E+000d 0.00E+000 1.0 - +7.6 9.40E-002
[ 4| 1] +0.00000000E+000 1.00000000E+000 3.03046132E+007 so Uin 1.00E+000d 0.00E+000 1.0 - +7.6 1.10E-001
Final values after iteration 4:
Final objective value ............. 0.0000000000E+000
Final constraint violation ........ 1.0000000000E+000
Final complementarity ............. 0.0000000000E+000 (0.0000000000E+000)
Final KKT conditions .............. 3.0304613191E+007 (0.0000000000E+000)
Unsuccessful termination: Minimum Stepsize reached in Armijo.
t_proc [s] t_wall [s] n_eval
nlp_f 0 0 2114
nlp_g 0 0 2120
nlp_grad_f 0 0 6
nlp_hess_l 0 0 5
nlp_jac_g 0 0 6
solver 0.156 0.156 1
As you can see MaxIter still the default one, but the same happens for all the other options that I am trying to tune.
The same happens in Matlab, which is 64bit; hence I would roule out a bitness issue.
Could be something related to the XML file in the lib folder? I tried to modify that file and also tried to delate it from the folder, but nothing changes.
Thank you!
Best,
Nicolò
Joris Gillis
Apr 17, 2018, 4:31:09 AM4/17/18
to CasADi
Dear Nicolo,
Could you test out this debug release ( https://github.com/casadi/binaries/releases/tag/commit-61cb2e5 ), with Worhp 1.12?
Thanks!
Best,
Joris
Could you test out this debug release ( https://github.com/casadi/binaries/releases/tag/commit-61cb2e5 ), with Worhp 1.12?
Thanks!
Best,
Joris
Nicolò Robuschi
Apr 17, 2018, 1:45:46 PM4/17/18
to CasADi
Dear Joris,
thanks for the effort!
unfortunately I have Py36 32bit on Windows, or Matlab 2016b, always on Windows. Could you provide me the binaries of one of these two languages (preferably Python)? Thanks!
Best,
Nicolò
thanks for the effort!
unfortunately I have Py36 32bit on Windows, or Matlab 2016b, always on Windows. Could you provide me the binaries of one of these two languages (preferably Python)? Thanks!
Best,
Nicolò
Joris Gillis
Apr 17, 2018, 4:14:20 PM4/17/18
to CasADi
Should appear in about half an hour..
Nicolò Robuschi
Apr 18, 2018, 4:04:32 AM4/18/18
to CasADi
Dear Joris,
thanks, now it works!
With this options:
opts_worhp = {}
opts_worhp['worhp.MaxIter'] = 10
thanks, now it works!
With this options:
opts_worhp = {}
opts_worhp['worhp.MaxIter'] = 10
opts_worhp['worhp.UserHM'] = False
opts_worhp['worhp.FidifHM'] = False
opts_worhp['worhp.BFGSmethod'] = 2
opts_worhp['worhp.Timeout'] = 60
solver = nlpsol('solver', 'worhp', nlp_prob, opts_worhp)
This is the output or WORHP 1.12
-------------------------------------------------------
This is WORHP 1.12-1, the European NLP-solver.
opts_worhp['worhp.Timeout'] = 60
solver = nlpsol('solver', 'worhp', nlp_prob, opts_worhp)
This is the output or WORHP 1.12
-------------------------------------------------------
This is WORHP 1.12-1, the European NLP-solver.
Use of WORHP is subject to terms and conditions.
Visit http://www.worhp.de for more information.
-------------------------------------------------------
Total number of variables ........................ 367
fixed variables 36
fixed variables 36
variables with lower bound only 0
variables with lower and upper bound 149
variables with upper bound only 0
Total number of box constraints .................. 334
Total number of other constraints ................ 274
equality constraints 274
Total number of other constraints ................ 274
equality constraints 274
inequality constraints with lower bound only 0
inequality constraints with lower and upper bound 0
inequality constraints with upper bound only 0
Gradient (user) 367/367 = 100.000%
Jacobian (user) 670/100558 = 0.666%
Block BFGS with intersections (1-3)
Algorithm Sequential Quadratic Programming
NLP MaxIter 10 Line Search Method Filter
Jacobian (user) 670/100558 = 0.666%
Block BFGS with intersections (1-3)
Algorithm Sequential Quadratic Programming
NLP MaxIter 10 Line Search Method Filter
QP MaxIter 500
LA solver MA97 (tol 1.0E-09, ref 10, ord METIS/AMD, scl none)
Tolerances:
Optimality (sKKT) 1.00E-06 (1.0E-03) IP ComTol 2.00E-07
Feasibility 1.00E-06 (1.0E-03) IP ResTol 4.00E-08
Complementarity 1.00E-06
Timeout 60.000 seconds
ITER OBJ CON OPTI/COMPL FLAGS ALPHA |DX| REL PEN REG TIME
[ 0| 4] +2.29053230E-001 2.76254591E-002 6.57310466E-002 so Bin 0.000E+000 3.85E-001 - - - 2.72E+000
[ 1| 2] +2.14317440E-001 2.37232105E-002 1.15159244E-002 sc Bin 1.000E+000 3.84E-001 0.85 - - 4.22E+000
[ 2| 3] +1.86686700E-001 1.41826062E-002 1.53575719E-002 so Bin 1.000E+000 1.13E+000 0.52 - - 5.64E+000
[ 3| 3] +1.67744423E-001 2.80021126E-003 1.74884458E-002 so Bin 1.000E+000 1.36E+000 0.14 - - 7.14E+000
[ 4| 10] +1.64495129E-001 7.14022537E-005 4.13698163E-004 so Baa 1.000E+000 3.07E-001 - - - 8.56E+000
[ 5| 2] +1.63509709E-001 2.25582490E-004 8.89790722E-004 so Baa 1.000E+000 3.52E-001 - - - 1.00E+001
[ 6| 7] +1.60844987E-001 9.68700470E-004 5.96488522E-003 so Ban 1.000E+000 9.79E-001 0.00 - - 1.15E+001
[ 7| 11] +1.52977802E-001 1.88273399E-003 1.34318522E-002 so Bin 1.000E+000 1.66E+000 0.00 - - 1.30E+001
[ 8| 11] +1.48499503E-001 8.30618357E-004 2.17588854E-002 so Ban 1.000E+000 1.04E+000 0.00 - - 1.44E+001
[ 9| 9] +1.43855667E-001 6.09263493E-004 2.07500916E-002 so Ban 1.000E+000 1.06E+000 0.00 - - 1.58E+001
[ 1| 2] +2.14317440E-001 2.37232105E-002 1.15159244E-002 sc Bin 1.000E+000 3.84E-001 0.85 - - 4.22E+000
[ 2| 3] +1.86686700E-001 1.41826062E-002 1.53575719E-002 so Bin 1.000E+000 1.13E+000 0.52 - - 5.64E+000
[ 3| 3] +1.67744423E-001 2.80021126E-003 1.74884458E-002 so Bin 1.000E+000 1.36E+000 0.14 - - 7.14E+000
[ 4| 10] +1.64495129E-001 7.14022537E-005 4.13698163E-004 so Baa 1.000E+000 3.07E-001 - - - 8.56E+000
[ 5| 2] +1.63509709E-001 2.25582490E-004 8.89790722E-004 so Baa 1.000E+000 3.52E-001 - - - 1.00E+001
[ 6| 7] +1.60844987E-001 9.68700470E-004 5.96488522E-003 so Ban 1.000E+000 9.79E-001 0.00 - - 1.15E+001
[ 7| 11] +1.52977802E-001 1.88273399E-003 1.34318522E-002 so Bin 1.000E+000 1.66E+000 0.00 - - 1.30E+001
[ 8| 11] +1.48499503E-001 8.30618357E-004 2.17588854E-002 so Ban 1.000E+000 1.04E+000 0.00 - - 1.44E+001
[ 9| 9] +1.43855667E-001 6.09263493E-004 2.07500916E-002 so Ban 1.000E+000 1.06E+000 0.00 - - 1.58E+001
ITER OBJ CON OPTI/COMPL FLAGS ALPHA |DX| REL PEN REG TIME
[ 10| 11] +1.39370740E-001 4.03624855E-004 2.57748984E-003 so Ban 1.000E+000 1.39E+000 0.00 - - 1.72E+001
Final values after iteration 4:
Final objective value ............. 1.6449512859E-001
Final constraint violation ........ 7.1402253656E-005
Final complementarity ............. 3.5433133675E-005 (3.5433133675E-005)
Final KKT conditions .............. 4.1369816298E-004 (3.8390942757E-003)
Terminating: Maximum number of iterations reached.
Successful termination: Acceptable Solution Found in previous iteration.
Final constraint violation ........ 7.1402253656E-005
Final complementarity ............. 3.5433133675E-005 (3.5433133675E-005)
Final KKT conditions .............. 4.1369816298E-004 (3.8390942757E-003)
Terminating: Maximum number of iterations reached.
Successful termination: Acceptable Solution Found in previous iteration.
t_proc [s] t_wall [s] n_eval
nlp_f 1.45 1.45 11
nlp_g 1.49 1.48 11
nlp_grad_f 5.86 5.86 12
nlp_jac_g 8.11 8.11 12
solver 17.3 17.3 1
Can I use this debug release instead of the 3.4.1 also with other NLP solvers, or is better to wait for an official 3.4.2 release?
Best,
Nicolò
nlp_g 1.49 1.48 11
nlp_grad_f 5.86 5.86 12
nlp_jac_g 8.11 8.11 12
solver 17.3 17.3 1
Can I use this debug release instead of the 3.4.1 also with other NLP solvers, or is better to wait for an official 3.4.2 release?
Best,
Nicolò
Fb
Jan 31, 2020, 12:23:39 PM1/31/20
to CasADi
Dear Joris,
I'm under ubuntu 18.04 using casadi 3.5.1, compiled with the WORHP option.
The WORHP solver is able to find the license but when I try to run any exemple as yours :
from casadi import *
x = MX.sym("x")
solver = nlpsol("solver","worhp",{"x":x,"g":x**2,"f":x**6},{"worhp": {"MaxIter":2}})
solver(lbg=1,ubg=2)
I get the following :
Opened license file /devel/license/worhp.lic
This module 'worhp' is licensed for versions <= 2.0 to:
bimec28
Université de Montréal
on host 'Bimec28' until 29.01.2021
This module 'worhp' is licensed for versions <= 2.0 to:
bimec28
Université de Montréal
on host 'Bimec28' until 29.01.2021
LICENSED FOR NON-PROFIT, ACADEMIC OR INSTRUCTIONAL USE.
YOU ARE NOT PERMITTED TO EXPLOIT THIS LICENSE COMMERCIALLY.
-------------------------------------------------------
Erreur de segmentation (core dumped)
Do you have an idea of how to solve this issue ? Would a backtrace from gdb help in anyway ?
Thanks in advance,
Francois
Reply all
Reply to author
Forward
0 new messages