shifted argument wouldn’t work with
multivariate exposure (see issue #175).mtp = FALSE wouldn’t work with
multivariate exposure.mtp argument has been changed from
FALSE to TRUE (see issue #170).lmtp_contrast()
that p-values aren’t adjusted for multiple comparisons (see issue
#172).lmtp_contrast() if
ref is a constant.onlySL = TRUE in predict.SuperLearner (see
issue #162).predict.SuperLearner to suppress some warnings.lmtp_survival() function for estimating the
entire survival curve. Enforces monotonicity using isotonic regression
(see issue #140).lmtp_survival() instead of the original values (see issue
#149).schoolmath which used a very slow
function for testing if a vector was “decimalish”.ipsi() for estimating
IPSI effects using the risk ratio.lmtp_control() now replaces extra estimator
arguments.intervention_type argument has been fully
deprecated..return_full_fits. Allows the user to
decide if full SuperLearner fit should be returned (issue #119).intervention_type argument replaced with
mtp.fits$id being NULL.
Fixes a backwards compatibility bug (issue #117).data.table version must be 1.13.0 or later. This was
when the function fcase was released (issue #122)..SL_folds argument split into
.learners_outcome_folds and
.learners_trt_folds.id with
lmtp_contrast (issue #110).folds must be greater than
1 (issue #112).shifted parameter for directly passing shifted data
instead of using a shift function (issue #89).intervention_type parameter required for specifying
if the intervention of interest is a static regime, a dynamic regime, or
a modified treatment policy (issue #94).return_all_ratios removed as an argument. Returned
density ratios are now non-cumulative product ratios.lmtp_tmle and lmtp_sdr
weren’t using validation set density ratios.data is a data.table
(issue #88).sim_point_surv data set (issue
#91).getting-started.Rmd vignette when using new version of the
future package (issue #100).data is a data.table
(issue #88).sim_point_surv data set (issue
#91).New weights parameter for observation sampling
weights (issue #78).
For time-to-event analysis, survival probability is now estimated instead of the cumulative incidence. This fixes a bug with IPW and survival problems.
Outcome type now accepts "survival" for explicit
indication of a survival outcome (issue #76). Because of this
lmtp_ipw() now requires setting the outcome type.
New .trimming parameter for trimming extreme density
ratios.
New .SL_folds parameter that controls the splits
used for fitting the SuperLearner (issue #84).
New .return_all_ratios parameter that allows for
returning non-cumulative product density ratios to the user.
bound parameter renamed to
.bound.
Fixed a bug that caused the final estimate to be incorrectly estimated with SDR (issue #87).
Fixed a bug that outputted outcome regressions and density ratios in incorrect order compared to the original data.
Fixed a bug in the missing data check that threw an error for missing data after an observation experiences the outcome.
Fixed a bug in the calculation of standard errors when the
id parameter is specified.
Fixed a bug that resulted in NA censoring indicators
throwing an error for missing data.
Fixed a bug about values() being deprecated in the
future package (issue #82).
Fixed a warning from the future package regarding random number generation (issue #81).
Fixed create_node_list() returns description (issue
#77).
slider dependency removed.
data.table added as a dependency.
event_locf() speed greatly improved (issue
#80).
Migrated continuous integration from Travis-CI to GitHub Actions.
Added a NEWS.md file to track changes to the
package.
License change to GPL-3.