Construct a single layer for use in an mb_seq()

A layer holds one or more modelblueprint objects that run in parallel. Each blueprint appends its prediction (under @yhat_name) to the dataset. aggregate_fn then receives the enriched dataset and can add further derived columns. yhat_name declares the layer's primary output column – this is what predict() returns when return_all = FALSE, and what downstream layers can reference as a feature.

Usage

mb_layer(blueprints, aggregate_fn = NULL, yhat_name = NULL)

Arguments

blueprints

A list of modelblueprint objects. Every blueprint must have @yhat_name set.

aggregate_fn

function(df) -> df. Receives the dataset after all blueprint predictions have been appended and returns it, optionally with additional columns. Defaults to the identity function when blueprints has one element; required when there are multiple blueprints.

yhat_name

[character(1)] The primary output column of this layer. When aggregate_fn is NULL, defaults to the single blueprint's @yhat_name. When aggregate_fn is provided, set this to the column aggregate_fn adds that represents the layer's combined prediction.

Value

An mb_layer object.

See also

mb_seq() to combine layers into a pipeline.

Examples

if (FALSE) { # \dontrun{
# Single model -- aggregate_fn and yhat_name default to the blueprint's yhat_name
l1 <- mb_layer(list(mb_freq))

# Two models combined as a product
l2 <- mb_layer(
  blueprints = list(mb_freq, mb_sev),
  yhat_name  = "pred_pure",
  aggregate_fn     = function(df) {
    df[["pred_pure"]] <- df[["pred_freq"]] * df[["pred_sev"]]
    df
  }
)
} # }