Hello, for some GPU intensive tasks I am using a cluster which uses LSF for job management. I can probably request all cards on a single machine (but that makes the job scheduler very unhappy), but it would certainly be helpful if I could either
use eg the bsub backend on cran for parallel point evaluation explicitly
expose a workflow like what you are probably doing for the parallel backends that allows me to: calculate and export the target parameters, do the calculations however I want, easily bind the results, update the gp. I think the only parts that are missing are a "calculate_only" argument to addIterations and a function to bind result arrays to an existing bayesOpt object
Hello, for some GPU intensive tasks I am using a cluster which uses LSF for job management. I can probably request all cards on a single machine (but that makes the job scheduler very unhappy), but it would certainly be helpful if I could either