Correct way to use a matrix in MILP objective function

[Zuumx]

Hi again, I'm having some trouble formatting the following as a MILP problem. I have 980 products and 4 groups. Each product has a attraction rate AR that differs for each group. Each product also has total sales. I want to maximize the sum of AR + Sales (normalized). I have the latest dev version.

The following MIPModel works just fine:

model <- MIPModel() %>%
  add_variable(x[p, g], p = 1:n, g = 1:m, type = "binary") %>%
  set_objective(sum_expr(ar[p, g] * x[p, g] +
                         total_sales[p] * x[p, g], p = 1:n, g = 1:m), "max") %>%
  # each product can be used once
  add_constraint(sum_expr(x[p, g], g = 1:m) <= 1, p = 1:n) %>% 
  # we want n_products selected for each group
  add_constraint(sum_expr(x[p, g], p = 1:n) == n_products, g = 1:m)

I'm bit confused how to model the ar part as MILPModel. The following works but yields different (wrong) results: (#238)

ar_vector <- ar[1:nrow(ar), 1:ncol(ar)] %>% unlist()

milpmodel <- MILPModel() %>%
  add_variable(x[p, g], p = 1:n, g = 1:m, type = "binary") %>%
  set_objective(sum_expr(ar_vector * x[p, g]  +
                         colwise(total_sales[p]) * x[p, g] , p = 1:n, g = 1:m), "max")
  # each product can be used once
  add_constraint(sum_expr(x[p, g], g = 1:m) <= 1, p = 1:n) %>% 
  # we want n_products selected for each group
  add_constraint(sum_expr(x[p, g], p = 1:n) == n_products, g = 1:m) 

I also tried the following:

arfunc <- function(p,g){
  ar_sub <- ar[p, g] 
  ar_vec <- as.vector(ar_sub) 
  return(ar_vec)
}

set_objective(sum_expr(w_ar * colwise(arfunc(p, g)) * x[p, g] +
                           w_sales * colwise(total_sales[p]) * x[p, g], p = 1:n, g = 1:m), "max") 

Error in as.data.frame.default(x[[i]], optional = TRUE) : 
  cannot coerce class ‘structure("LinearVariableCollection", package = "ompr")’ to a data.frame

What is the correct way to format this problem?

[Zuumx]

Nevermind, I changed the function to:

afunc <- function(p, g) {
  vapply(seq_along(p), function(k) ar[p[k], g[k]], numeric(1L))
}

And the objective function is now correct. Both functions give same results but I guess MILPModel handles them differently.