parameter tolerance reached

[Brent Morgan]

Hello,

I have many parameters (thousands) and they all have a small range, so small differences make a change in the solution.

I am having to re-run ceres over and over, using the optimal solution from the previous run for the current run, to find a most optimal solution.  Do I change the parameter tolerance here?  Thanks,

sample output for d, g, h, s

d: 0.00e+00 g: 9.05e-03 h: 0.00e+00 s: 0.00e+00 e:  0 

d: 1.18e-04 g: 5.25e-02 h: 8.38e-03 s: 3.24e-01 e:  2 

d: 1.54e-04 g: 3.22e-02 h: 9.72e-03 s: 5.35e-02 e:  1 

d: 2.26e-04 g: 2.49e-02 h: 1.43e-02 s: 1.05e-01 e:  1 

d: 2.65e-04 g: 1.35e-02 h: 2.38e-02 s: 1.29e-01 e:  1 

d: 2.68e-04 g: 1.08e-02 h: 3.39e-02 s: 2.46e-01 e:  1 

d: 2.04e-04 g: 2.31e-02 h: 4.09e-02 s: 2.67e-01 e:  1 

d: 1.32e-04 g: 2.60e-02 h: 3.48e-02 s: 1.68e-01 e:  1 

d: 1.50e-04 g: 3.41e-02 h: 7.36e-02 s: 2.98e-01 e:  3 

d: 4.36e-05 g: 3.39e-02 h: 3.17e-02 s: 1.33e-01 e:  1 

d: 4.33e-05 g: 3.50e-02 h: 9.26e-02 s: 3.11e-01 e:  2 

d: 4.24e-06 g: 3.38e-02 h: 2.96e-02 s: 9.66e-02 e: 20 

d: 1.28e-05 g: 3.13e-02 h: 4.34e-02 s: 1.10e-01 e: 13 

d: 1.20e-05 g: 2.94e-02 h: 4.02e-02 s: 8.66e-02 e: 20 

d: 1.29e-05 g: 2.73e-02 h: 4.22e-02 s: 7.86e-02 e: 20 

d: 1.11e-05 g: 2.54e-02 h: 4.09e-02 s: 6.82e-02 e: 13 

d: 9.18e-06 g: 2.36e-02 h: 3.84e-02 s: 5.82e-02 e: 20 

d: 7.25e-06 g: 2.21e-02 h: 3.62e-02 s: 5.05e-02 e: 20 

d: 5.64e-06 g: 2.06e-02 h: 3.31e-02 s: 4.26e-02 e: 20 

d: 3.96e-06 g: 1.92e-02 h: 3.82e-02 s: 4.56e-02 e: 20 

d: 1.75e-06 g: 1.81e-02 h: 2.71e-02 s: 3.04e-02 e: 20 

d: 1.61e-06 g: 1.74e-02 h: 1.94e-02 s: 2.02e-02 e: 20 

d: 1.89e-06 g: 1.60e-02 h: 3.82e-02 s: 3.66e-02 e: 20 

d: 9.73e-09 g: 1.60e-02 h: 4.51e-03 s: 4.06e-03 e: 20 

d: 1.61e-05 g: 6.05e-03 h: 2.21e-02 s: 1.21e-02 e:  5 

Best,

Brent

[Sameer Agarwal]

Brent,

What is the reason the solver is terminating every time? can you share the output of Summary::FullReport from a couple of the solves?

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/10bb3a34-715f-4545-96f6-af7568861254%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Brent Morgan]

Sameer,

Parameters                               6000

Line search direction              LBFGS (20)

Line search type                  CUBIC WOLFE

Minimizer iterations                       11

Time (in seconds):

  Cost evaluation                      0.0000 (0)

  Gradient evaluation                 94.9815 (25)

  Polynomial minimization              0.0009

Total                                 95.0048

Termination:                      CONVERGENCE (Parameter tolerance reached. Relative step_norm: 0.000000e+00 <= 1.000000e-08.)

Is this helpful?

Best,

Brent

[Sameer Agarwal]

Brent,This looks like the search direction is becoming useless/getting, and you are restarting the LBFGS approximation gets it going again.
You may want to reduce lbfgs rank.
there is also use_approximate_eigenvalue_bfgs_scaling that may help.

Sameer

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/99a2d6d7-e751-4d5e-adc4-26e947b62060%40googlegroups.com.