Back to the main page.
Bug 1421 - test_bug1309 ends with SVD error on NaN
| Status | CLOSED WORKSFORME |
| Reported | 2012-04-12 17:39:00 +0200 |
| Modified | 2019-08-10 11:56:10 +0200 |
| Product: | FieldTrip |
| Component: | inverse |
| Version: | unspecified |
| Hardware: | PC |
| Operating System: | Linux |
| Importance: | P3 normal |
| Assigned to: | |
| URL: | |
| Tags: | |
| Depends on: | |
| Blocks: | |
| See also: |
Boris Reuderink - 2012-04-12 17:39:51 +0200
To replicate: >> test_bug1309 Output of revision 6510: scanning grid location 35696/35699 scanning grid location 35697/35699 scanning grid location 35698/35699 scanning grid location 35699/35699 found minimum after scanning on grid point [-30 -21 -51] First-order Iteration Func-count f(x) Step-size optimality 0 4 NaN NaN Warning: dipole lies on boundary of volume model > In forward/private/inf_medium_leadfield at 54 In forward/private/eeg_leadfieldb at 93 In ft_compute_leadfield at 386 In dipole_fit>dipfit_error at 230 In optim/private/lineSearch>bracketingPhase at 103 In optim/private/lineSearch at 48 In optim/private/fminusub at 208 In fminunc at 367 In dipole_fit at 146 In ft_dipolefitting at 437 In test_bug1309 at 27 Error using ==> svd Input to SVD must not contain NaN or Inf. the call to "ft_dipolefitting" took 266 seconds and an estimated 0 MB
Boris Reuderink - 2012-06-04 15:48:31 +0200
Note to self: apparently, to problem occurs in dipole_fit; and it seems that numerical optimisation finds gradient that contains NaNs --- deep inside MathWorks optimization routines.
Boris Reuderink - 2012-06-05 17:50:22 +0200
Note: It appears that the finite-difference method used in the numerical optimization procedure causes a gradient with NaNs. These NaN subsequently spread to other variables.
Boris Reuderink - 2012-11-02 13:29:57 +0100
I am no longer working on FieldTrip. Hence, I donate all my bugs to the joint development user.
Jan-Mathijs Schoffelen - 2012-11-07 13:47:46 +0100
according to the dashboard the test passes