Runs an optimization, using multiple starting values, in an attempt to find a global minimum. Can take advantage of mpi4py as long as the start-parameter-objects can be pickled (and subsequently sent to remote MPI processes).
The self.rootRank master MPI process uses asynchronous communication to hand out individual (single-start) optimization jobs to remote worker MPI processes.
Methods
| __init__(singleStartOptimizer[, rootRank, ...]) | Initialise. |
| broadcast(prmIterator) | |
| broadcastResultList() | |
| doAsyncMpiMasterSends() | |
| doAsyncMpiMultiStart() | |
| doAsyncMpiWorkerRecvs() | |
| doEvenDivisionMultiStart() | |
| minimize(startPrmIterator) | |
| optimize(startPrmIterator) | |
| pickleToByteArray(obj) | |
| setNumMpiProcessesForAsyncProcessing(numMpiProcs) | Set the number of MPI processes which triggers the use of asynchronous master/slave processing of the multi-start optimizations. |
| unpickleFromByteArray(ba) |