Can I presolve an ILP model before optimization?

问题

I am getting out of memory while building the model. Is there any way, to reduce the model while building it using existing functions?

Details: Assume I have the following model (from the docs here section Presolve. The real code also uses sparse matrices, so this is just to figure out what can be done further):

min 2*x1 - 5*x2 + 3*x3 + 10*x4
s.t.
x1 + x2 + x3 = 15 (1)
x1 <= 7           (2)
x2 <= 3           (3)
x3 <= 5           (4)
x4 > 1            (5)

Clearly the only way that all of these constraints can be satisfied is if x1 = 7, x2 = 3, and x3 = 5. My goal is to reduce dimensions "on the fly" if possible. In pseudo-code:

model <- build_model(objective_function,
                     restrictions (1) to (4))
model1 <- presolve_model(model)
model2 <- build_model(objective_function1,
                     restrictions model1 and (5))
result <- gurobi::gurobi(model2)

Where model1 only consists of the variable x4 as x1 = 7, x2 = 3, and x3 = 5 (presolved). Is this possible?

Comments:

In Gurobi's Python interface you can perhaps use presolve.model()? See here but I have no clue how that is done. I also didn't find a possibility to return the presolved model from gurobi::gurobi(). However, the last two lines in the reprducible example return the model as a file - but NOT the presolved, as can be seen from the example.
Gurobi does presolving, as can be seen from the parameter Presolve.
Experts might want to have a look at this package.
Maybe it is related to the vbasis and cbasis argument from Gurobi? The docs state

Finally, if the final solution is a basic solution (computed by simplex), then vbasis and cbasis will be present.

Reproducible example:

model <- list()
model$A          <- matrix(c(1, 1, 1, 0, 
                             1, 0, 0, 0, 
                             0, 1, 0, 0, 
                             0, 0, 1, 0, 
                             0, 0, 0, 1), nrow = 5, ncol = 4, byrow = T)
model$obj        <- c(2, -5, 3, 10)
model$modelsense <- "min"
model$rhs        <- c(15, 7, 3, 5, 1)
model$sense      <- c('=', '<=', '<=', '<=', '>')
model$vtype      <- 'I'
params <- list(OutputFlag = 1, Presolve = 2, TimeLimit = 3600)

result <- gurobi::gurobi(model, params) # optimize

# gurobi::gurobi_write(model, 'mymodel.mps') # output to file
# gurobi::gurobi_write(model, 'mymodel.lp') # output to file

回答1:

It is not possible with v8.0. As @GregGlockner stated:

As stated in this stackoverflow questions "Gurobi lets you access the presolved model, but only from the Python API"

来源：https://stackoverflow.com/questions/49734762/can-i-presolve-an-ilp-model-before-optimization

标签

gurobi