Comparative denchmark of https://github.com/asherikov/qpmad QP solver
Alexander Sherikov
e9d46300e2
qpmad 1.2.0
|
vor 2 Jahren | |
|---|---|---|
| ariles @ 4eb1e3e1eb | vor 4 Jahren | |
| cmake | vor 4 Jahren | |
| eiquadprog @ 3c6006c9bd | vor 2 Jahren | |
| figures | vor 2 Jahren | |
| qpOASES @ 326a6517da | vor 4 Jahren | |
| qp_collection @ 91526180bf | vor 4 Jahren | |
| qpmad @ 65924534a7 | vor 2 Jahren | |
| test | vor 2 Jahren | |
| .gitignore | vor 4 Jahren | |
| .gitmodules | vor 4 Jahren | |
| CMakeLists.txt | vor 4 Jahren | |
| Makefile | vor 2 Jahren | |
| README.md | vor 4 Jahren |
README.md
Comparative benchmark of QP solvers
Solvers:
qpmad-> https://github.com/asherikov/qpmadqpOASES-> https://github.com/coin-or/qpOASESeiquadprog-fast-> https://github.com/stack-of-tasks/eiquadprog
Results
- All problems are dense and positive-definite.
qpOASESis used with MPC option preset, this gives ~50% performance boost compared to default preset.- Time measurements include only the time required to solve a problem; other operations, such as initialization of the solver or reshaping of the constraints, are not taken into account.
oneshot (qpOASES)
The problem set includes 'standalone' problems distributed with qpOASES -> https://github.com/coin-or/qpOASES/tree/misc/testingdata/cpp/problems, positive-semidefinite problems are skipped.
Hotstarting is not used.
Time (second) required to solve each of the problems:
The same plot with logarithmic scale:
iterative (qpOASES)
The problem set includes MPC problems distributed with qpOASES -> https://github.com/coin-or/qpOASES/tree/misc/testingdata/cpp/oqp, all of them have constant Hessian and varying constraints.
qpOASESis hotstarted starting from the second problem in the sequence, factorization of the Hessian is reused byqpmadin the same way.eiQuadProgdoes not produce the exact answer (up to 1e-9 precision) on some of the problems.
Plots correspond to specific MPC problems and demonstrate time (second) taken
by the solvers on each MPC iteration:
