Package index • modelblueprint

The modelblueprint object

Create, predict from, persist, and manipulate a modelblueprint. A modelblueprint wraps a fitted model together with its training data, pipeline functions, and deployment metadata.

modelblueprint: modelblueprint: a model-agnostic container for ML model lifecycles
predict(<modelblueprint>): Generate predictions from a modelblueprint
savemb() saveMB(): Save a modelblueprint to disk
loadmb() loadMB(): Load a modelblueprint from disk

Accessors and updaters

Tidy extract_* and set_* verbs for reading and replacing individual slots. Each set_* returns a new modelblueprint — never mutates in place — and runs the S7 validator automatically.

extract_fit(): Extract the fitted model from a modelblueprint
extract_train() extract_test() extract_holdout(): Extract a data split from a modelblueprint
extract_pre_process_fun() extract_feat_eng_fun() extract_post_process_fun(): Extract pipeline functions from a modelblueprint
extract_original_inputs() extract_feature_names(): Extract feature name vectors from a modelblueprint
extract_target() extract_yhat_name() extract_exposure_name() extract_exposure_value() extract_exposure_zero_rep() extract_offset_name() extract_offset_value() extract_display_name() extract_deploy_notes(): Extract scalar metadata from a modelblueprint
set_model(): Swap the fitted model inside a modelblueprint
set_train() set_test() set_holdout(): Replace a data split in a modelblueprint
set_pre_process_fun() set_feat_eng_fun() set_post_process_fun(): Replace pipeline functions in a modelblueprint
set_original_inputs() set_feature_names(): Replace feature name vectors in a modelblueprint
set_target() set_yhat_name() set_exposure_name() set_exposure_value() set_exposure_zero_rep() set_offset_name() set_offset_value() set_display_name() set_deploy_notes(): Set scalar metadata on a modelblueprint

Data manipulation

dplyr-style verbs that operate on all data splits inside a modelblueprint simultaneously, returning a new modelblueprint.

filter(<modelblueprint>): Filter rows in a modelblueprint's datasets
mutate(<modelblueprint>): Mutate columns in a modelblueprint's datasets
left_join(<modelblueprint>): Left-join into a modelblueprint's datasets

Model sequences

Chain multiple modelblueprints into a sequential pipeline where the output of one model feeds the input of the next.

mb_seq(): A sequential pipeline of model layers
mb_layer(): Construct a single layer for use in an mb_seq()
predict(<mb_seq>): Generate predictions from an mb_seq

Validation and diagnostics

Quantify model performance and calibration. All functions accept either a modelblueprint (uses the stored data and model) or a plain data frame (bring your own predictions).

gain(): Cumulative Gains Chart
gain(<default>): Cumulative Gains Chart (default method)
pred_vs_obs(): Predicted vs Observed Calibration Plot
residuals_grouped(): Grouped Residuals vs Predicted Plot

Feature analysis

Understand how individual features relate to the target and drive model predictions.

one_way(): Create a one-way analysis plot
one_way(<modelblueprint>): One-way analysis for a modelblueprint
distribution(): Plot the distribution of the target variable
distribution(<modelblueprint>): Target distribution for a modelblueprint
pdp(): Partial dependence plot for any predict()-compatible model
pdp(<modelblueprint>): Partial dependence plot for a modelblueprint
shap(): SHAP Feature Importance and Dependence Plots
shap(<modelblueprint>): SHAP plots for a modelblueprint
sami(): SAMI Double Lift Chart

Batch validation

Generate and save a full suite of validation, one-way, PDP, stability, and SHAP plots for every dataset split in a single call.

model_validation(): Generate and save model validation plots

Dashboard

Interactive Shiny app that combines all diagnostics and feature analysis tools into a single four-tab interface.

mb_dashboard(): Launch an interactive dashboard for a modelblueprint

Utilities

Helper functions useful in pre- and post-processing pipelines.

unitise(): unitise a numeric variable to the range 0 to 1
save_plots(): Save plots or HTML widgets to a single HTML file

Example modelblueprints

Ready-made modelblueprint objects for exploring the package without needing your own data. Includes regression, classification, Poisson frequency (car insurance), random forest, XGBoost, and H2O examples.

mb_lm_regression() mb_lm_classification() mb_glm_regression() mb_glm_binomial() mb_glm_poisson() mb_glm_poisson_freq() mb_rpart_regression() mb_rpart_classification() mb_rf_regression() mb_rf_classification() mb_xgb_regression() mb_xgb_classification() mb_h2o_regression() mb_h2o_classification() mb_h2o_glm_large(): Example modelblueprint objects