• Modify the package structure
• Collect features from formula

• Adding warning message for upcoming changes.

Added

• Extra step to check consistency ofdelta_n with exposure range.
• Software paper examples were added.

Changed

• Plotting pseudo population includes object details. Setinclude_details = TRUE.
  *generate_pseudo_pop does not takeY as an input.

Changed

• Docker image supports R 4.2.3
• generate_syn_data supportsvectorized_y to accelerate data generation.
• matching_fun -->dist_measure
• matching_l1 -->matching_fn
• estimate_semipmetric_erf now takes thegam models optional arguments.
• estimate_pmetric_erf now takes thegnm models optional arguments.
• trim_quantiles -->exposure_trim_qtls
• generate_pseudo_pop function acceptsgps_obj as an optional input.
• internal_use is not part of parameters forestimate_gps function.
• estimate_gps function only returnsid,w, and computedgps as part of dataset.
• Now the design and analysis phases are explicitly separated.
• gps_model -->gps_density. Now it takes,normal andkernel options instead ofparametric andnon-parametric options.

Added

• estimate_npmetric_erf supports bothlocpol andKernSmooth approaches.
• There isgps_trim_qtls input parameter to trim data samples based on gps values.
• Now users can also collect the original data in the pseudo population object.

Fixed

• A bug with swapping transformed covairates with original one.

Fixed the failing unit tests due to the following bug report:
https://bugs.r-project.org/show_bug.cgi?id=18337

• Fixed regression tests that happened due to changes in wCorr.
• Droppedoptimized_compile flag.

In this version upgrade we:

• Dropped importingKernSmooth andtidyr packages.
• Droppedpred_model argument. The package only uses SuperLearner for prediction models.
• Added features to use a more optimized algorithm for a commonly used simplified case (scale = 1).
• Added effective sample size.
• Added Kolmogorov-Smirnov (KS) statistics for the generated pseudo-population (usesEcume package).
• Madesl_lib a required argument.
• Removedearth andranger packages from mandatory imports.
• Standardized the trimming approach to be less confusing for the users.
• Modified internal kernel smoothing approach.
• Renamed a couple of internal parameters for clarity and uniformity in the package.
• Fixed a bug on the covariate balance threshold.

Fixed

• Message for not implemented methods changed to reduce misunderstanding.
• Empty counter will raise error in estimating non-parameteric response function.

Changed

• matching_l1 returns frequency table instead of entire vector.
• Vectorized population compilation and used data.table for multi-thread assignment.
• Removed nested parallelism in compiling pseudo population, which results in close control on memory.
• estimate_npmetric_erf also returns optimal h and risk values.

Added

• estimate_gps returns the optimal hyperparameters.
• estimate_gps returns S3 object.
• Internal xgboost approach supportverbose parameter.
• Pseudo-population object now report the parameters that are used for the best covariate balance.

Fixed

• Naming covariate balance scores.

Changed

• Restarting adaptive approach to keep trying up to maximum attempt.

Added

• Synthetic data (synthetic_us_2010)
• Check on not defined covariate balance (absolute_corr_fun, absolute_weighted_corr_fun)
• Covariate balance threshold type: mean, median, maximal.
• Improved test coverage.
• Singularity definition file.

Added

• added the status of optimized compile to generate_psuedo_pop function output.
• compute_closest_wgps accepts the number of user-defined threads.

Changed

• Vignette file names.
• The trim condition from > and < into >= and <=.
• Removed seed input from generate_syn_data function. In R package, setting seed value inside function is not recommended. Users can set the seed before using the function.
• OpenMP uses user defined number of cores.

Fixed

• Initial covariate balance for weighted approach. The counter column was not preallocated correctly.
• Counter value for compiling. The initial value was set to one, which, however, zero is the correct one.
• Private variable issue with OpenMP.
• Fixed OpenMP option on macOS checks.