Get computed incremental step

[Jesús Briales García]

Hi,

I'm trying to debug some optimization and was interested in seeing the incremental step computed by the solver in each iteration.

The iteration summary provides the step norm, but not the whole vector. Is it possible to access this somehow?

Best regards,

- Jesus Briales

[Sameer Agarwal]

Hi Jesus,

No short of putting some debugging statements yourself in the code, there isn't a way to access it. 

What is it that you are trying to do?

it is possible to dump the gauss-newton system to disk though and you could load it in matlab and compute the step yourself if youlike.

http://ceres-solver.org/nnls_solving.html#_CPPv2N5ceres6Solver7Options41trust_region_minimizer_iterations_to_dumpE 

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/9bad50ac-f492-4636-a1e5-1b5620cf44c5%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Jesús Briales García]

Hi Sameer,

Thanks for your answer! Actually I was looking at alternatives to figure out where some convergence issues where coming from, but got that fixed for now.

Also the ability to dump the Gauss-Newton system seems really useful, I'll keep that in mind!

Best regards,

- Jesus Briales

[Keir Mierle]

Hi Jesús,

Depending on what you mean by "incremental step", you can get at the intermediate parameter values for successful steps. In particular, if you set Solver::Options::update_state_every_iteration, then your parameter blocks will be updated before the IterationCallback is invoked. If you want to see what has changed, you can cache the parameter values and compare.

You can use this mechanism to, for example, plot what the parameters are doing during the optimization.

Happy optimizing,

Keir

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/0e541982-3e8d-4549-8dc6-8b2ae8b180d1%40googlegroups.com.

[Jesús Briales García]

Hi Keir,

That approach also sounds good.

Actually my question arose because I assumed that the result of solving the linear system, before applying this step onto the state, would be available somewhere. From your answers I understand that's not the case but the proposed workarounds sound well enough :)

Thanks again!

- Jesus