interactions: "Quadratic expression are not supported"

[seschaub]

dear all,

first of all, thank you Drik for creating and maintaining the package.

Today, I was trying to solve a model with "interaction terms". However, I received an error message: "Quadratic expression are not supported"

Bellow, you find a simplified version of the model. In this model the interaction is "* (1 + x[i])".
Does someone know how to solve this? do i simply miss something?

define parameters

soil_coef <- c(2, -1)
manage_coef <- c(0.5, -.1)

number of plots

n = 100

information about land properties

dz <- data.frame(
soil = runif(n, min = 5, max = 10))
cov.mat_me <- dz %>% as.matrix()

model definition

model_1 <- MIPModel() %>%

define variable that we optimize over

add_variable(x[i], i = 1:n, type = "binary") |>

set objective function

set_objective(
sum_over(
# output 1
(as.numeric(cov.mat_me[i,1]) * soil_coef[1] + x[i] * manage_coef[1] ) * (1 + x[i]) +
# output 2
(as.numeric(cov.mat_me[i,1]) * soil_coef[1] + x[i] * manage_coef[1]) * (1 + x[i]),
i = 1:n),
"max") |>

add the constraint

add_constraint(sum_over(x[i], i = 1:n) <= 5 ) %>%

solve_model(with_ROI(solver = "glpk")) |>
get_solution(x[i])
model_1

[benwatsoncpa]

Hi, first off, big thank you for developing this library. I'm running into the same issue but it may be that I'm setting up something incorrectly.

Here is a reproducible example:

`x <- c(1, 2, 3, 4, 5)
y <- c(3, 9)

m <- length(x)
n <- length(y)

model <- MIPModel() %>%
add_variable(x[i],i=1:m,type="integer") %>%
add_variable(y[j],j=1:n,type="integer") %>%
add_variable(d[i,j],i=1:m,j=1:n,type="binary") %>%
set_objective(1,sense="min") %>%
add_constraint(sum_over(d[i,j],j=1:n)<=1,i=1:m) %>%
add_constraint(sum_over(d[i,j]*x[i],i=1:m)==y[j],j=1:n) `

Update: I reformulated my code to work, seems that I didn't need to add x & y:

X <- c(1, 2, 3, 4, 5)
Y <- c(9,5,0)

x <- length(X)
y <- length(Y)

model <- MIPModel() %>%
add_variable(D[x,y],x=1:x,y=1:y,type="binary") %>%
set_objective(1,sense="min") %>%
add_constraint(sum_over(D[x,y],y=1:y)<=1,x=1:x) %>%
add_constraint(sum_over(D[x,y]*X[x],x=1:x)==Y[y],y=1:y) %>%
solve_model(with_ROI("glpk", verbose = TRUE))

[sbmack]

The ompr documentation does not suggest a non-linear optimization capability. However the R ROI package allows the formulation of non-linear models: http://roi.r-forge.r-project.org/index.html

[dirkschumacher]

Yes quadratic terms are not supported. It wouldn't be too much work though to add them at this point. Right now though I don't have a lot of time. The infrastructure is there. rmpk supports it.