NEWS.md

In mixgb: Multiple Imputation Through 'XGBoost'

mixgb 2.2.3

Changes

• data_clean() now becomes check_data(). This function performs preliminary sanity check before imputation. A few checks have been added.
• Some arguments' names have changed in show_var() function for compatibility with vismi package.
• The dataset nhanes3_newborn has been replaced with newborn, with variables renamed for clarity and to align with the vismi package.
• Variable names in nhanes3 have been updated to maintain consistency with newborn.

mixgb 2.1.3

New Features

• When save.models = TRUE in mixgb(), the return object now includes the mean and variance of imputed values for each variable for each iteration. This would allow users to plot convergence diagnostics using package vismi.

mixgb 2.0.3

Bug fix

• global variable warning resolved

mixgb 2.0.2

Deprecation

• Removed support for bootstrap in mixgb()

Related to XGBoost Parameter Changes

• related to xgb.train():
• num_class now passed through params list, need to set num_class=NULL for non-multiclass imputation.

Refactorisation

• Refactored mixgb_null() function for better readability and maintainability - using helper functions in impute_each.R. Other relevant functions will be updated later.

mixgb 2.0.1

Bug fix

• Removed macOS-specific -framework Accelerate flag for Linux builds in Makevars file

related to XGBoost Parameter Changes

• related to xgb.train():
• num_class now passed through params list.
• remove reshape = TRUE argument in predict() function as it is deprecated in XGBoost >= 2.0.0.

mixgb 2.0.0

Breaking Changes

• The package is now compatible with XGBoost version 3.1.1.1 or higher.

Related to XGBoost Parameter Changes

• related to xgb.train():
• watchlist is changed to evals

• objective and eval_metric are now passed through params list.

• related to xgb.cv():

• convert data from sparse.model.matrix into xgb.DMatrix format. Note, this is a quick fix for minimal safe change, and we plan to further optimise data handling in future releases.
• best.nrounds is now obtained via cv.train$early_stop$best_iteration instead of cv.train$best_iteration.

mixgb 1.5.3

Bug fix

• Fix an error in cbind2(Mis.m, Obs.m) that occurred when the imputed dataset had only a single incomplete variable.
• Cause of error: In this scenario, Mis.m (a matrix of all other incomplete variables except the currently imputed one) was a 0x0 matrix, which caused error during binding.

mixgb 1.5.2

For CRAN Submission

• Significantly faster imputation by optimising data preprocessing and the use of RcppArmadillo
• Visual diagnostic functions have been moved to the vismi package
• Update Documentation

mixgb 1.4.2

Bug fix

• Set drop.unused.levels = FALSE in fac2sparse() to prevent dropping unused levels in factor or ordinal factor.
  • Ensure the feature matrix has the same number of columns to feed in XGBoost

mixgb 1.4.1

Compatibility

• Compatible with both XGBoost >=2.0.0 or CRAN version of XGBoost (=1.7.5.1)

Bug fix

• Correct feature type for numeric and integer variable for initial imputation
• Update save_yhatobs() for Type 1 pmm.

mixgb 1.4.0

• Optimised mixgb() for large datasets:
• Significantly faster imputation by optimising data preprocessing and the use of RcppArmadillo
• Enhanced memory efficiency with in-place modifications using data.table
• Bootstrapping option removed from mixgb(). Users can still use bootstrap in the archived function mixgb0().
• PMM is now set to NULL by default.

mixgb 1.3.2

Miscellaneous

• Improves package documentation regarding the import of xgb.save() and xgb.load() from XGBoost.

mixgb 1.3.1

Compatibility

• Makes the package compatible with XGBoost 2.0.0 with GPU support:
• Introduces a new parameter device.
• Deprecates parameters gpu_id and predictor.
• Sets tree_method = "hist" by default, aligning with XGBoost 2.0.0.

New Features

• Introduces support for saving imputation models to a local directory through the parameter save.models.folder in mixgb().
• Models save in JSON format using xgb.save(), a method recommended by XGBoost for future compatibility.
• When save.models.folder is specified, the return object of mixgb() includes the current imputed datasets, directories for imputation models, and relevant parameters. This object can save using saveRDS() as it doesn't directly contain the models. Users can later load this object into R and employ impute_new(object, newdata, ...) for new data imputation.

mixgb 1.2.1

Updates

• Includes the URL of the published article.

mixgb 1.2.0

New Features

• Enhances mixgb(data,...) to support datasets with diverse data types:
• numeric
• integer
• factor
• logical

Note: Users must manually convert character variables to factors.

mixgb 1.1.0

New Features

• Introduces default_params(), an auxiliary function for mixgb(), to validate the list of XGBoost hyperparameters supplied by the user. It simplifies hyperparameter modifications without requiring explicit specification of all default values.

Bug Fixes

• Addresses issues related to PMM for multiclass variables.
• Updates plot_hist() and plot_bar() to align with changes in ggplot2 3.4.0:
• Replaces ..density.. with after_stat(density) in plot_hist().
• Replaces ..prop.. with after_stat(prop) in plot_bar().

mixgb 1.0.2

Minor Changes

• Adjusts examples to use nthread = 2 to comply with CRAN policies.

mixgb 1.0.1

Changes in Default Settings

• Transitions from bootstrapping to subsampling. Subsampling, set at subsample = 0.7, becomes the default method due to identified biases with bootstrapping in certain scenarios.
• Default for mixgb():
  • Subsampling: subsample = 0.7.
  • No bootstrapping: bootstrap = FALSE.

mixgb 0.1.1

Minor Bug Fixes and Updates

• Resolves a minor issue in the createNA() function.
• Modifies default settings in mixgb():
• ordinalAsInteger: Changes from TRUE to FALSE.
• max_depth: Changes from 6 to 3.
• nrounds: Changes from 50 to 100.
• bootstrap: Sets to TRUE by default.

mixgb 0.1.0

Initial Release

• First version releases on CRAN.
• Supports both single and multiple imputation.
• Offers customisable settings for bootstrapping and predictive matching.
• Provides visual diagnostics for multiply imputed data.

Notes

• Mac OSX users might face challenges with multithreading in mixgb as xgboost requires OpenMP for multi-core operations. For details, please refer to OpenMP for Mac.

mixgb documentation built on Jan. 17, 2026, 5:07 p.m.