The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.
bag_tree() with the "rpart" engine no
longer accepts case weights. Fit the model without case weights instead
(#390).
bag_tree() with the "rpart" engine now
passes tree_depth, min_n, and
cost_complexity on to the trees, so default fits now use
the bag_tree() defaults of cost_complexity = 0
and min_n = 2. To restore the previous fits, set
cost_complexity = 0.01 and min_n = 20
(#406).
boost_tree() with the "mboost" engine
no longer accepts case weights because mboost doesn’t predict from a
weighted fit. Fit the model without case weights instead
(#363).
boost_tree() with the "mboost" engine
no longer allows setting nu via set_engine().
Use learn_rate instead (#395).
boost_tree() with the "mboost" engine
now warns and ignores control and
tree_controls when set via set_engine().
Previously they silently overrode the main arguments, such as
trees and tree_depth. Use the main arguments
of boost_tree() instead (#414).
survival_prob_*() functions now error when given the
defunct time argument. Use eval_time instead
(#419).
survival_prob_coxph() no longer has the defunct
x argument. Use object instead
(#419).
boost_tree() with the "mboost" engine
now passes learn_rate on to
mboost::boost_control(nu), making it tunable. Previously it
was silently ignored (#395).
multi_predict() is now available for
boost_tree() with the "mboost" engine over the
trees submodel parameter (#290).
null_model() gains a "survival" engine
for censored regression, fitting a Kaplan-Meier curve via
survival::survfit() (#353).
rand_forest() gains a "randomForestSRC"
engine for censored regression, fitting a survival random forest via
randomForestSRC::rfsrc() (#130).
rand_forest() gains a "ranger" engine
for censored regression, fitting a survival random forest via
ranger::ranger() (#131).
bag_tree() with the "rpart" engine now
passes tree_depth, min_n, and
cost_complexity, as well as other
rpart::rpart.control() arguments set via
set_engine(), on to the trees. Previously they were
silently ignored (#406).
decision_tree() with the "rpart" engine
now correctly returns the median survival time of the leaf’s
Kaplan-Meier curve for type = "time" predictions, instead
of rpart’s relative event rate (#331).
proportional_hazards() with the
"glmnet" engine now predicts on data with factors when
fitted through fit_xy() (#365).
proportional_hazards() with the
"survival" engine now predicts when a single
strata() term contains more than one variable,
e.g. strata(s1, s2) (#404).
survival_prob_*() and hazard_*() now
validate their inputs and return more informative error messages when
given an unusable object, new_data, or
eval_time (#271).
survival_reg() with the "survival"
engine and strata() terms now returns NA
survival and hazard predictions for new data rows with a missing value
in a stratification variable, instead of silently using another
stratum’s scale (#383).
Engine-specific tuning parameters are now registered by censored instead of parsnip (#417).
eval_time = Inf are now not always set to 0 and confidence
intervals at infinite evaluation times are now not always set to
NA. This applies to proportional_hazards()and
bag_tree() models as well as models with the
partykit engine, decision_tree() and
rand_forest() (#320).predict() methods for flexsurv
models, in preparation for the upcoming flexsurv release (#317).multi_predict() is now available for all prediction
types for proportional_hazards() models with the
"glmnet" engine, so newly also for
type = "time" and type = "raw" (#277,
#282).
Random forests with the "aorsf" engine can now
predict survival time, i.e., predict(type = "time") is now
available (#308).
survival_prob_*(), survival_time_*(),
and hazard_*() helper functions now all take a parsnip
model_fit object as the main input, instead of an engine
fit as was the case for some of them previously (#302).extract_fit_engine() now works properly for
proportional hazards models fitted with the "glmnet" engine
(#266).
multi_predict(type = "survival") for
proportional_hazards(engine = "glmnet") models: when used
with a single penalty value, this value is now included in
the results. It was previously omitted (#267, #282).
proportional_hazards(engine = "glmnet") models now
don’t pretend to be able to deal with sparse matrices when they are not
(#291).
Fixed a bug for
proportional_hazards(engine = "glmnet") where prediction
didn’t work for a workflow() with a formula as the
preprocessor (#264).
survival_time_coxnet() and
survival_prob_coxnet() gain a multi argument
to allow multiple values for penalty (#278, #279).The new eval_time argument replaces the
time argument for the time points at which to predict
survival probability and hazard. The time argument has been
deprecated (#244).
The matrix interface for fitting, fit_xy(), now
works for censored regression models (#225, #234, #247, #251).
Improved error messages throughout the package (#248).
Added the new "aorsf" engine for
rand_forest() for accelerated oblique random survival
forests with the aorsf package (@bcjaeger, #211).
Added the new flexsurvspline engine for
survival_reg() (@mattwarkentin, #213).
Predictions of type "linear_pred" for
survival_reg(engine = "flexsurv") are now on the correct
scale for distributions where the natural scale and the unrestricted
scale of the location parameter are identical,
e.g. dist = "lnorm" (#229).
Predictions of type "linear_pred" for
proportional_hazards(engine = "glmnet") via
multi_predict() now have the same sign as those via
predict() (#242).
Predictions of survival probability for
survival_reg(engine = "flexsurv") for a single time point
are now nested correctly (#254).
Predictions of survival probability for
decision_tree(engine = "rpart") for a single observation
now work (#256).
Predictions of type "quantile" for
survival_reg(engine = "survival") for a single observation
now work (#257).
Fixed a bug for printing coxnet models, i.e.,
proportional_hazards() models fitted with the
"glmnet" engine (#249).
Predictions of survival probabilities are now calculated via
summary.survfit() for proportional_hazards()
models with the "survival" and "glmnet"
engines, bag_tree() models with the "rpart"
engine, decision_tree() models with the
"partykit" engines, as well as rand_forest()
models with the "partykit" engine (#221, #224).
Added internal survfit_summary_*() helper functions
(#216).
For boosted trees with the "mboost" engine, survival
probabilities can now be predicted for time = -Inf. This is
always 1. For time = Inf this now predicts a survival
probability of 0 (#215).
Updated tests on model arguments and update()
methods (#208).
Internal re-organisation of code (#206, 209).
Added a NEWS.md file to track changes to the
package.
These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.