ck37r: Chris Kennedy's R Toolkit

News for ck37r

impute_missing_values
initial support for GLRM imputation
prauc_table
table of precision-recall AUCs for SL or CV.SL ensembles.

doMC is no longer supported.

auc_table - default null hypothesis is now the highest AUC.
csvl_auc_table() - table of AUCs for each learner, plus Discrete SL and SL.
factors_to_indicators
return indicators as integers rather than numerics.
significant speed improvements
add max_levels limit, default to 200.
allow missing values in factors
h2o_init_multinode - create a multi-node h2o cluster e.g. on a SLURM system.
impute_missing_values
add all_vars argument; if T save the imputation values even when variable is not missing data. This allows new datasets to be imputed to the values from the training data. Treated labelled columns as numerics.
support integer64 numeric type (data.table)
impute_missing_values
treat haven::labelled values as numerics.
save variable names into the impute list.
support supplying imputation values from another dataset, for use when scoring a model to a larger dataset.
load_all_code() - changed default environment from baseenv() to .GlobalEnv, which will make explicit package references unnecessary.
missingness_indicators - return matrix of integers rather than numerics.
set_java_memory() - specify maximum memory usage for rJava JVM.
get_java_memory() - get maximum memory allocated to a rJava JVM.
Mode - option to not choose NA as a mode, set TRUE by default.
plot_roc - plot the best learner, not necessarily the SuperLearner.
sl_h2o_auto - automatic machine learning via h2o.
sl_mgcv - add mgcv wrapper for splines.
sl_xgboost_cv - integrate support for gaussian family outcomes.
vim_corr - new function to rank covariates by their (possibly weighted) univariate correlation with the outcome.

factors_to_indicators
don't drop matrix to a vector, e.g. when using analyzing a single covariate.
impute_missing_values
fixed important bug where incorrect column indices could be used when skip_vars was not blank.
fix missingness indicators when no indicators need to be created.
missingness_indicators
Fix when only one column of missingness indicators needs to be created.
plot_roc - fix axis labels for updated ggplot2.