Ceres solver crash in general unconstrained minimization

[Aditya Gadre]

Hi,

I am using general unconstrained minimization setup to smooth out a set of data points. The cost function involves computing approximate tangent angles at each data point. For computing cost function derivatives, I am using NumericDiffFirstOrderFunction. I ran into a crash with the following error message (Ceres bug ... contact the developers!). It looks like the gradient had a value of inf at bracket_low point.

low_rank_inverse_hessian.cc:92 Skipping L-BFGS Update, delta_x_dot_delta_gradient too small: -50.1724, tolerance: 1e-14 (Secant condition).
line_search.cc:705 Check failed: bracket_low.value_is_valid && bracket_low.gradient_is_valid Ceres bug: f_low input to Wolfe Zoom invalid, please contact the developers!, initial_position: [x: 0.00000000e+00, value: 1.15368273e+02, gradient: -4.07593559e+03, value_is_valid: 1, gradient_is_valid: 1], bracket_low: [x: 5.78229541e-04, value: 1.13917515e+02, gradient: inf, value_is_valid: 1, gradient_is_valid: 0], bracket_high: [x: 0.00000000e+00, value: 1.15368273e+02, gradient: -4.07593559e+03, value_is_valid: 1, gradient_is_valid: 1]

From some online posts, I gathered that using sqrt function in the cost computation may be problematic. I don't know if that is what caused this error. I am unable to reproduce this, but will try.

I will appreciate any inputs about what this error means!

Thank you very much for your help!

- Aditya

[Sameer Agarwal]

This should not happen. I thought we had a guard against this sort of thing.

It would be good to have a reproduction of this problem though to make sure that it is indeed a ceres bug.

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 visit https://groups.google.com/d/msgid/ceres-solver/3091f97a-6399-493e-a89e-763f09ed1572n%40googlegroups.com.