The code given shows how Weka algorithms can be run in parallel across distributed computers using Matlab to take advantage of available hardware and quicken algorithm completion time on multiple data sets or parameter sets.
Parallel processing of Weka algorithms is handled here in Matlab as support for parallel processing in Weka does not seem to be fully fledged as yet.
A timeout can be specified on algorithm/job processing time. This is useful if algorithms can be potentially long-running or non-terminating, or when there are a large number of experiments to ensure they finish within a reasonable amount of time.
Before the provided code can be executed:
* Define a parallel configuration in Matlab. See "Configuring Parallel Processing.txt".
* Assign the configuration name to variable "config" in runParallelWeka.m.
* Copy the Weka library weka.jar to the folder ParallelWeka. The jar file can be found in Weka's program directory.
* Copy the folder ParallelWeka to the same location on all worker machines.
Matlab must be installed on all machines hosting workers.
To run the code, add the path of the folder ParallelWeka to Matlab's paths on the local machine, and at the command line enter "runParallelWeka".
Products required updated.