Grouped Residuals vs Predicted Plot — residuals

Bins predictions by exposure, computes grouped residuals, and overlays a loess trend line with a 95% confidence interval. Useful for diagnosing systematic model bias across the prediction range.

Usage

residuals_grouped(data, ...)

# Default S3 method
residuals_grouped(
  data,
  pred = "predict",
  obs = "observed",
  exposure = "exposure",
  exposure_per_bin = 10,
  residual_type = c("raw", "pearson"),
  title = "",
  ret = c("plot", "data"),
  ...
)

# S3 method for class 'modelblueprint'
residuals_grouped(
  data,
  set = c("train", "test", "holdout"),
  exposure_per_bin = 2500,
  residual_type = c("raw", "pearson"),
  title = NULL,
  ret = c("plot", "data"),
  ...,
  precomputed_preds = NULL
)

Arguments

data: A modelblueprint object.
...: Passed to residuals_grouped.default().
pred: [character(1)] Name of the predictions column.
obs: [character(1)] Name of the observed target column.
exposure: [character(1)] Name of the exposure column. If the column is absent, every row is given weight 1. Default "exposure".
exposure_per_bin: [numeric(1)] Target exposure per bin. Default 2500. Automatically reduced if the dataset is too small for meaningful grouping.
residual_type: [character(1)] "raw" or "pearson".
title: [character(1)] Chart title. Defaults to model_display_name (with the set name appended when plotting multiple sets).
ret: [character(1)] "plot" or "data".
set: [character] Dataset splits to use: any of "train", "test", "holdout". Defaults to all available (non-NULL) sets. When more than one set is used, a named list with one result per set is returned.
precomputed_preds: [numeric | NULL] Optional vector of pre-computed predictions (one per row of the requested set). When supplied, the internal predict.modelblueprint() call is skipped.

Value

A plotly object or data.table depending on ret.

Examples

# \donttest{
df <- data.frame(
  obs      = rbinom(500, 1, 0.3),
  pred     = runif(500, 0.1, 0.5),
  exposure = rep(1, 500)
)
residuals_grouped(df, pred = "pred", obs = "obs", exposure = "exposure")

# }
# \donttest{
mb <- modelblueprint(
  model  = glm(vs ~ wt + hp, data = mtcars, family = binomial),
  train  = mtcars,
  y_name = "vs",
  model_display_name = "logistic_vs"
)
residuals_grouped(mb)

# }