Skip to contents

Launch an interactive dashboard for a modelblueprint

Source: R/dashboard.R
mb_dashboard.Rd

Opens a Shiny app that provides an interactive view of a fitted modelblueprint. The app has four tabs:

Usage

mb_dashboard(mb, ...)

Arguments

mb

A modelblueprint object. Must have at least one of @train, @test, or @holdout set.

...

Currently unused. Reserved for future arguments.

Value

A shiny.appobj. The app launches in the browser when the return value is printed (the normal behaviour when called interactively). Returns invisibly when assigned.

Details

  • Summary – model class, display name, target, exposure, dataset row counts, sum of target and exposure per split, the full variable list, and an overlaid density chart of the target vs model predictions.

  • Validation – gain chart, predicted vs observed calibration, and grouped residuals. All three can be shown side-by-side across train, test, and holdout sets simultaneously, making overfitting immediately visible.

  • PDPs – partial dependence plot for any variable in @x_original_inputs, with controls for bins, aggregation strategy, and sample size. Aggregated data can be downloaded as CSV.

  • One-ways – exposure-weighted mean of the target across bins of any feature, with an optional model prediction overlay and split variable. Aggregated data can be downloaded as CSV.

All sidebar controls (bins, aggregation type, dataset) update plots reactively. Errors in individual plots are caught and displayed inline so a single failing chart never crashes the app.

Required packages

shiny, bslib, and plotly must be installed. These are listed under Suggests so they are not installed automatically with the package. Install them with:

install.packages(c("shiny", "bslib", "plotly"))

Install shinycssloaders for loading spinners on slow plots:

install.packages("shinycssloaders")

See also

pdp(), one_way(), pred_vs_obs(), gain(), residuals_grouped() for the underlying functions used inside the dashboard.

Examples

if (FALSE) { # \dontrun{
mb <- modelblueprint(
  model              = glm(vs ~ wt + hp, data = mtcars, family = binomial),
  train              = mtcars,
  y_name             = "vs",
  x_original_inputs  = c("wt", "hp"),
  model_display_name = "logistic_vs"
)

mb_dashboard(mb)
} # }