quicopt

The R client for the Quicopt optimization service.

You describe a decision: what you get to choose, what has to hold, and what you want as much (or as little) of as possible. Quicopt finds the best choice there is.

Part of that decision is often taken before the data is known — demand, prices, yields. Declare that data as random variables, and the model is solved over scenarios drawn from their distributions: minimize an expectation or a conditional value at risk, and require constraints to hold with a given probability. Observed history in a data.frame becomes a stochastic program in one call.

There is no solver on your machine; the service does the solving.

Install

install.packages("quicopt", repos = c("https://quicopt.r-universe.dev",
                                      "https://cloud.r-project.org"))

Or straight from the source repository — the package is pure R, so this needs no compiler either:

pak::pak("Quicopt/quicopt-r")

Toy Model

library(quicopt)

m <- model()
x <- num_var(m, "x", 0, 4)
y <- bin_var(m, "y")
maximize(m, 3 * x + 5 * y)
add(m, x + 2 * y <= 5)

res <- solve(m)
res$status; res$objective   # "optimal" 14
res$solution

Optimization under Uncertainty

Order x units at 3 apiece against a demand you will only learn later, pay for the mismatch, and meet demand in at least 90% of scenarios:

m <- model()
x <- num_var(m, "x", 0, 200)                     # decide now
demand <- rand_var(m, "demand", normal(100, 15)) # learn later
set_scenarios(m, 512, seed = 42)

minimize(m, 3 * x + 10 * expectation(max(demand - x, 0)))
add(m, prob(demand - x <= 0) >= 0.9)

res <- solve(m)
res$solution[["x"]]                              # 118.7: the 90% service
                                                 # level binds at the sample q90

max(demand - x, 0) is the shortfall — and it is plain R: arithmetic, max, sum and friends build the model's expressions directly. Note that scenarios are drawn by the service from the model's own seed (set_scenarios), so set.seed() plays no role here.

Have the uncertainty as data instead of a distribution? Every column of a data frame becomes a random variable, jointly, with correlation preserved:

history <- read.csv("demand_price.csv")   # observed rows
m <- model()
set_empirical(m, history)                 # columns -> sources, nrow -> scenarios

License

Apache License 2.0 — see LICENSE. (c) 2026 Tim Bode, PGI-12, Forschungszentrum Jülich.