Improve MATLAB integration

[MLopez-Ibanez]

Currently, communication with MATLAB is slow, requires editing R files and quite cumbersome and fragile. How to fix this?

1. Figure out what is the best way to communicate to MATLAB the information that irace generates. Currently, we use:

  matlab_function_call <- "Main(INSTANCE,SEED,A,B)"
  args <- paste0(names(configuration), "=", configuration, collapse=";")
  exitcode <- matlabr::run_matlab_code(
                         # This is the instance and seed
                         c(paste0("INSTANCE='", instance, "';SEED=", seed),
                           # The configuration parameters,
                           args, matlab_function_call),
                         paths_to_add=matlab_script_path,
                         verbose= (debugLevel >= 2))

which uses the matlabr package to run the code:

INSTANCE='instance'
SEED=seed
A=value_of_A
B=value_of_B
Main(INSTANCE,SEED,A,B)

The above assumes that there are 2 parameters A and B. If the parameters are different, then the above code must be edited.

This is not very flexible, because the user has to create such a function Main() with the precise parameters, but I don't know what would be a better way. Is there a way to call a MATLAB function with a variable number of parameters?

There is another package: https://cran.r-project.org/web/packages/R.matlab/index.html

I think using R.matlab package to send/receive data from/to MATLAB would be much better than using matlabr.

2. Figure out what is the best way to collect the output given by MATLAB. Currently, we read the output like:

  ## Parse the output (just a single number!)
  # FIXME: Create a uniquely named temporary file instead of Result.txt to
  # allow parallel execution
  output <- scan(file="Result.txt", quiet=TRUE)
  cost <- as.numeric(output)

That is the function Main() in MATLAB must save the output to a file Result.txt, which irace reads. The above is slow and very unflexible. Ideally we would be able to call MATLAB and get the output directly without any extra files, but I don't know how to do that. Maybe possible with the R.matlab package mentioned above.

3. Once the above is figured out, create an R package tunematlab that contains a function: irace_matlab_runner(). And in scenario.txt you would just specify "targetRunner="matlab" and irace will load the tunematlab package and use the irace_matlab_runner() function to call matlab in the right way. This is the easy part.