NEWS
LRDE 1.1.2
- Improved numerical stability of Hurdle-NB parameter estimation.
- Added safeguards against non-finite likelihood values during
BFGS optimization.
- Improved handling of extreme negative binomial parameters and
zero-truncation probabilities.
- Added protection against zero or non-finite initial parameter
values.
- Improved initialization of group-specific mean parameters in
tag-wise dispersion estimation.
- Added probability clipping to prevent infinite logit
transformations.
- Improved robustness of tag-wise dispersion estimation for sparse
genes and genes with all-zero counts.