Skip to contents

Optimization function using HiGHS solver

Source: R/solvers.R
HiGHS_sol_com.Rd

Solves the integer/linear programming problem for computing the minimum test statistic in stratified randomized experiments using the HiGHS solver.

Usage

HiGHS_sol_com(
  Z,
  block,
  weight,
  coeflists,
  p,
  ms_list,
  exact = TRUE,
  block.sum = NULL
)

Arguments

Z

An n-dimensional treatment assignment vector.

block

An n-dimensional vector specifying block of each unit.

weight

A B-dimensional vector of block weights.

coeflists

A list of H elements, each containing B matrices with block-specific test statistics for different values of k.

p

Upper bound on the number of units with effects greater than c.

ms_list

A list containing mu (means) and sigma (standard deviations) for each test statistic.

exact

Logical; if TRUE, solve as integer linear program (ILP). If FALSE, solve as linear program (LP relaxation).

block.sum

Optional pre-computed block summary from summary_block().

Value

A list with:

sol

Solution vector

obj

Optimal objective value

Details

This function formulates and solves an optimization problem to find the minimum value of the combined test statistic under the constraint that at most p units have effects greater than a threshold c.

The problem has the following structure:

  • Variables: x_sj (binary/continuous), eta_h, theta

  • Objective: minimize theta

  • Constraints: probability constraints per stratum, coverage constraint, test statistic equality constraints, and normalization constraints

See also

Gurobi_sol_com for the Gurobi implementation, solve_optimization for the unified wrapper

Examples

if (FALSE) { # \dontrun{
# This is typically called internally by pval_comb_block()
# Direct usage requires pre-computed coefficient lists
result <- HiGHS_sol_com(Z, block, weight, coeflists, p, ms_list, exact = TRUE)
} # }