run_missing_glucose_imputation() now handles both
explicit missing glucose values and missing readings implied by
timestamp gaps. When timestamps skip expected CGM intervals, the
function regularizes each subject to the expected interval and imputes
the newly created missing glucose rows.
The returned data frame is now simpler. It contains the user’s
original columns plus imputed_glucose_value. Internal
columns used for timestamp regularization, lag features, rolling means,
model fitting, and missingness tracking are no longer returned.
The original glucose column is still preserved. Values that were
originally missing, or created from timestamp gaps, remain
NA in the original target column, while completed values
are stored in imputed_glucose_value.
imputed_glucose_value is returned as a continuous
numeric model estimate. Users who need whole-number glucose values for
reporting can round this column after imputation.
run_missing_glucose_imputation() now supports
selectable real-imputation methods through the existing
models argument. The default models = "auto"
keeps the missing-rate rule, using MICE+ARIMA when
missingness is at or below the configured threshold and
MICE+XGBoost otherwise.
Users can now force one final real-imputation method with
models = "arima", "xgboost",
"rf", "knn", or "lightgbm".
Random Forest, kNN, and LightGBM use the same lag-feature workflow as
the existing ARIMA and XGBoost real-imputation paths.
Real-imputation model engines now use n_threads = 1
by default for CRAN-friendly and shared-system-friendly CPU use. Users
can increase n_threads for faster local XGBoost, Random
Forest, or LightGBM runs.
Added a bundled Shiny app for interactive missing glucose imputation. The app lets users upload a CSV file or load example data, choose the relevant columns, select the final imputation method, run imputation, preview results, and download the completed data.
Added built-in example data for demonstrating both explicit missing glucose values and timestamp-gap handling.
The optional Python-compatible backend remains available with
imputer_backend = "sklearn". The default backend remains
imputer_backend = "mice" for standard R usage. Both
backends support the selectable final imputation methods, with Python
LightGBM available when the optional Python lightgbm module
is installed.
Updated README and vignettes to describe timestamp-gap handling, the simplified output structure, selectable final imputation methods, the bundled Shiny app, backend options, and post-imputation rounding.