flatten_ml_fit_problem: Return a flattened representation of a multi-level fitting problem instance

Description

This function transforms a multi-level fitting problem to a representation more suitable for applying the algorithms: A matrix with one row per controlled attribute and one column per household, a weight vector with one weight per household, and a control vector.

Usage

flatten_ml_fit_problem(
  ml_problem,
  model_matrix_type = c("combined", "separate"),
  verbose = FALSE
)
as_flat_ml_fit_problem(x, model_matrix_type = c("combined", "separate"), ...)

Arguments

ml_problem

A fitting problem created by ml_problem() or returned by flatten_ml_fit_problem().

model_matrix_type

Which model matrix building strategy to use? See details.

verbose

If TRUE, print diagnostic output.

An object

...

Further parameters passed to the algorithm

Value

An object of classes flat_ml_fit_problem, essentially a named list.

Details

The standard way to build a model matrix (model_matrix = "combined") is to include intercepts and avoid repeating redundant attributes. A simpler model matrix specification, available via model_matrix = "separate", is suggested by Ye et al. (2009) and required for the ml_fit_ipu() implementation: Here, simply one column per target value is used, which results in a larger model matrix if more than one control is given.

Examples

Run this code

# NOT RUN {
path <- toy_example("Tiny")
flat_problem <- flatten_ml_fit_problem(ml_problem = readRDS(path))
flat_problem

fit <- ml_fit_dss(flat_problem)
fit$flat_weights
fit$weights
# }