Enforcing and relaxing parameter bounds

[Milan Vukov]

Hello,

I am solving a parameter estimation problem, where, for validation purposes I want to enforce bounds in the parameters: if at the solution a bound is hit, that should most certainly mean that either problem setup is ill posed and/or a dataset is somehow corrupted. The bounds are set a valid range: e.g. I expect the parameter to be in the range +-0.01, but I set the bounds to be +-0.1.The initial guess I provide to the solver is always feasible, e.g. 0 in the example I gave.

For some tests everything is OK, and the bounds are respected. For some other test problems, I ran into issue of slow convergence -- this may be a problem of scaling and/or the fact the solver is not allowed to make large steps any more in some directions. Moreover, the solution inspection reveals that bounded parameters are far away from the bounds...

Now is tricky part: I relaxed bounds to be +-1e6, but I don't get the same solution as if there were no bounds (my starting point). This seems weird. Any idea why this is happening?

Cheers,

Milan

[Milan Vukov]

Probably closely related to this one: https://groups.google.com/forum/?fromgroups#!topic/ceres-solver/SPuRbreNvcI

[Sameer Agarwal]

Milan,

Yes most likely this is related to https://github.com/ceres-solver/ceres-solver/issues/245

though it would help if I could take a look at your problem and what the solver is doing while solving it. Is it possible for you to share a sample problem with me?

Sameer

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/d2ebebcd-aeab-4477-982a-ec3fd562c932%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.