Predictions via multi_predict() for glmnet models
now correctly default to mean counts instead of the linear predictor
(#89).
Predictions for a single observation now work for
poisson_reg() with the "glmnet" engine
(#48).
Predictions of type "conf_int" for the
"stan" engine now use the new function suggested by
rstanarm (#86).
Moved imports and declaration of global variables into the standard place (#59).
Tests on the model specification have been updated to the current testing pattern in parsnip and other extension packages (#55).
Tests are now self-contained (#60).
Removed obsolete check on supplying newdata as an
argument to predict() or multi_predict()
(#87).
The predict() method for the "glmnet"
engine now checks the penalty value via
parsnip::.check_glmnet_penalty_predict() instead of a
copied version of the function (#57).
Removed the now obsolete registration of the
predict_raw() generic as it is now exported from parsnip
(#52).
Removed the unused internal s3_register() function
(#53).
poisson_reg()) were
moved to the parsnip package.A default engine of glm was added for
poisson_reg().
Added tidy() methods for pscl::hurdle()
and pscl::zeroinfl().
Work-around for a glmnet bug where different column
order will silently produce incorrect predictions.
multi_predict() was enabled.
Updates to go along with new version of
parsnip.
NEWS.md file to track changes to the
package.