It is very helpful to show which parameters (collectively) affect the R-factor the strongest and weakest. Those can be obtained using svd of the covatiance matrix:
U,s,V=svd(covariance_matrix)
Note however, the strangeness here:
V[0], V[1] - are the ones which affect the R-factor the least.
V[N-2],V[N-1] - are the ones which affect the R-factor the most.
V[N],V[N+1] - are the ones which DO NOT affect the R-factor (if exist).
This discontinuity is probably due to the nature of pseudo-inverse used to find the covariance matrix in levmar, and should be detected probably by the appropriate value of the eigenvalue.
It is very helpful to show which parameters (collectively) affect the R-factor the strongest and weakest. Those can be obtained using svd of the covatiance matrix:
Note however, the strangeness here:
V[0], V[1]- are the ones which affect the R-factor the least.V[N-2],V[N-1]- are the ones which affect the R-factor the most.V[N],V[N+1]- are the ones which DO NOT affect the R-factor (if exist).This discontinuity is probably due to the nature of pseudo-inverse used to find the covariance matrix in levmar, and should be detected probably by the appropriate value of the eigenvalue.