- All functions were re-written, by adapting from stats:: and new S3 methods when required
- Uses R's internal C code for all matrix decompositions, and LAPACK when possible
- Fixes rounding problem with broom's exponentiation
- Performs better on Windows, the previous code relied heavily on Unix-specific matrix products
- Allows
family = gaussian
, etc instead offamily = gaussian()
, just likeglm()
- Supports
sandwich::vcovBS()
- Supports
subset = (cyl == 4)
and similar subsetting
- Adds generic method for weights (i.e. now I can replicate the clotting example from glm)
- Starts adding vignettes with all lm/glm tests from R source code
- Allows to use formulas such as
mpg ~ log(wt)
instead of having to createmtcars$logwt <- log(mtcars$wt)
- Implements some fixes from speedglm package to solve estimation bugs with categorical variables
- Adds the subset argument for a closer match with the stats package
- Makes all the arguments similar to the closest degree to lm() and glm()
- This is the first version, which basically fits my need to obtain GLM (Quasi-Poisson) estimates