- For distributed computing using MPI
- Requires Rmpi package with MPI environment
- In Windows, use MS-MPI
- in Linux, openmpi is recommended while intel mpi or mvapich works too
- Comparison of regular exhaustive search of original FSelector package with doSNOW/doMC/Rmpi implementation
- doSNOW and doMC works in multicore environment within a single node
- Rmpi version will work in a single or multiple nodes
- The current implementation uses Iris test from FSelector documentation, and may not work in very large number of MPI ranks
- Comparison of regular greedy search of original FSelector package with Rmpi implementation
- The current implementation uses Iris test from FSelector documentation, and may not work in very large number of MPI ranks
- In addition to splitting loops over multiple ranks, polling version is provided, as a sample work of dynamic load balancing