Paralleling in Octave

[Maria Jose Casas Serrano]

Hi,

It’s the first time I’m trying to parallelise some code and I’m stuck in a basic concept that I hope someone could help me with.

 
First, I’ll try to explain what my code is about. I’m doing Evolutionary Computation and I have a large population of elements that represent the genomes of modelled plants. This population should be evaluated by a fitness function and then, some of the elements will evolve depending on the probabilities of some genetic operators.

The bottleneck happens when evaluating them so, I thought that I could parallelise that part of the code. 

A simplified example (with a population of 10 elements) is the following:

P =

{

  [1,1] = G

  [1,2] = +[-G]GG

  [1,3] = +[[GG]G]

  [1,4] = G-[-GG]

  [1,5] = GG+GG[-GG]

  [1,6] = []

  [1,7] = [-G]GGG

  [1,8] = [--[[G]]]

  [1,9] = +G+GGG

  [1,10] = [[GG]]

}

mytest.m 

inputparam2 = 0.01;

inputparam3 = 0.2;

logger.initSLF4O();

numCores = nproc();

timeStart = tic();

fun = @(idx) evaluationParallel(P{idx,:}, inputparam2, inputparam3);

[outputparam1, outputparam2, outputparam3] = pararrayfun(numCores - 1, fun, rows(P), "UniformOutput", false);

elapsed_time = toc(timeStart);

logger.info("mytest: Elapsed time %d", elapsed_time);

-----------------------------------------------------------------------------------------------------------------------------------

evaluationParallel.m

function [outputparam1, outputparam2, outputparam3] = evaluationParallel(individual, inputparam2, inputparam3)

logger.initSLF4O();

 # TODO: Remove it. Only for testing reasons

logger.info("evaluationParallel: individual %s", individual);

logger.info("evaluationParallel: inputparam2 %f", inputparam2);

logger.info("evaluationParallel: inputparam3 %f", inputparam3);

# END TODO

logger.info("evaluationParallel: Building the phenotype of the individual %s", individual);

[outputparam3{individual}, height, width, branchesnumber] = LsystemGenerator(individual);

logger.info("evaluationParallel: Calculating the aspect ratio of the individual %s", individual);

[outputparam1(:,individual), outputparam2(:,individual)] = getAspectRatio(height, width, branchesnumber, inputparam2, inputparam3);

endfunction

---------------------------------------------------------------------------------------------------------------------------------

However in the “inputparam2” and “inputparam3” parameters, the evaluationParallel function is receinving the second and third elements of the population.

Do you know what I am doing wrong? Perhaps I misunderstood the parallelisation concept but I thought I could apply it in a piece of code meanwhile there weren’t dependencies with the rest of the process.

Thanks in advance

Cheers

Maria J