Value for best_x is worse than for x about ecos HOT 4 CLOSED

best_x is only an internal variable holding the "best iterate" encountered so far, and is used to backtrack in case the search direction computation has numerical issues, and no further progress can be made. You should NOT use this variable - use x instead (I think the documentation does not mention x_best at all because it's an internal variable) as defined by the interface in ecos.h.

best_x is only updated if certain criteria are met, and AFTER checking that the exit conditions for the current iterate are not met. With which exitflag does your instance terminate?

from ecos.

Thanks for the clarifications! Yes, I get exitflag 0, so it is optimal. Do you have a complete example of how to use ECOS from C, that shows the necessary steps to get the answer and do error and optimality checking? In the examples/ directory I found only two Python examples.

from ecos.

The best example is probably runecos.c in the src directory, or the mex interface ecosmex.c in the matlab directory, as well as all tests in the test directory. You need your matrix data in column compressed storage format. The rest is pretty straightforward - See also the interface ecos.h. There's also a section in on C in the readme.

On Jul 8, 2014, at 0:57, alexvoronov [email protected] wrote:

Thanks for the clarifications! Yes, I get exitflag 0, so it is optimal. Do you have a complete example of how to use ECOS from C, that shows the necessary steps to get the answer and do error and optimality checking? In the examples/ directory I found only two Python examples.

—
Reply to this email directly or view it on GitHub.

from ecos.

Thanks!
It was actually ecos.h file that tricked me into using best_x :)
I read through ecos_mex.c and see that exitflags are just forwarded back to matlab. File runecos.c also do mostly timing and does do much about the answer. But I think I figured out by now enough to run ECOS. I also use it with QCML, so I rely on QCML for column compressed storage.

from ecos.