
2026-11-15 Steven E. Pav <shabbychef@gmail.com>
* 0.1.5 on CRAN
* Fix gradient of harsmlik.
* Fix harsmfit computation of etahat, erank, R2 and delta_R2 when offset is given.
* Fix predict.linodds when newdata on missing factor levels. #8
* predict.linodds saves na.action attribute of the model frame.
* erank checks for mu summing to one, errors out otherwise.
* erank can deal with zero probabilities.
* harsmlik and hensmlik throw errors on empty input.
* hensmfit now stores the estimate, not the maxLik object, in coefficients.
* fix linodds class attribute of hensm, harsm.
* hensmfit checks length of beta0.
* harsmfit names coefficients from the column names of X, like hensmfit.
* sanity checks for reg_wt, reg_power, and so on.
* fix regularization gradient around zero.
* default reg_zero in harsm and harsmfit is now NULL, as in hensm.
* rhenery throws error on NA and NAN in mu.
* smax returns NA, rather than a zero, for NA input when values overflow.
* tidy and glance methods for harsm and hensm return visibly.
* documentation fixes.
* finite difference checks of all analytic gradients into unit tests.

2025-09.07 Steven E. Pav <shabbychef@gmail.com>
* 0.1.4 on CRAN
* Adding regularization. #9
* Adding hensmfit experts function. #10
* Adding broom tidy and glance methods. #11

2025-04-19 Steven E. Pav <shabbychef@gmail.com>
* 0.1.3 on CRAN
* Fix issue with factor levels in predict. #8

2024-10-25 Steven E. Pav <shabbychef@gmail.com>
* 0.1.2 on CRAN
* Allow warm start in harsm and hensm.

2019-10-14 Steven E. Pav <shabbychef@gmail.com>
* 0.1.1 on CRAN
* Change default in harsm and hensm to use unnormalized weights,
correcting inference when not all finishes are observed.

2019-10-01 Steven E. Pav <shabbychef@gmail.com>
* 0.1.0 on CRAN

2018-09-17 Steven E. Pav <shabbychef@gmail.com>
* start work

