Emrys::FitOptions< Scalar > Struct Template Reference

Options for fitMixture(). More...

#include <EmrysFilter/Distributions/Fitting.h>

Public Attributes
int	maxIterations = 100
	Maximum number of EM iterations (E-steps).
Scalar	tolerance = Scalar(1e-8)
	Stop when an iteration improves the weighted log-likelihood by at most this.
Scalar	regularization = Scalar(1e-6)
	Covariance floor, relative to the source's variance in each dimension: every component covariance gets \( \epsilon \Sigma_{ii} \) added to its \( i \)-th variance.
Scalar	minLogWeight = Scalar(-20)
	Components whose normalized log-weight falls below this are dropped.

Detailed Description

template<typename Scalar>
struct Emrys::FitOptions< Scalar >

Options for fitMixture().

Member Data Documentation

regularization

template<typename Scalar>

Scalar Emrys::FitOptions< Scalar >::regularization = Scalar(1e-6)

Covariance floor, relative to the source's variance in each dimension: every component covariance gets \( \epsilon \Sigma_{ii} \) added to its \( i \)-th variance.

Scaling each dimension by its own variance keeps the floor consistent when the dimensions have different units. Keeps components that capture duplicate particles from collapsing.

---

The documentation for this struct was generated from the following file:

• EmrysFilter/Distributions/Fitting.h