Implement Cross-Validation Using Parallel Computing

Simple Parallel Cross Validation

In this example, use crossval to compute a cross-validation estimate of mean-squared error for a regression model. Run the computations in parallel.

mypool = parpool()
Starting parpool using the 'local' profile ... connected to 2 workers.

mypool = 

  Pool with properties:

    AttachedFiles: {0x1 cell}
       NumWorkers: 2
      IdleTimeout: 30
          Cluster: [1x1 parallel.cluster.Local]
     RequestQueue: [1x1 parallel.RequestQueue]
      SpmdEnabled: 1

opts = statset('UseParallel',true);

load('fisheriris');
y = meas(:,1);
X = [ones(size(y,1),1),meas(:,2:4)];
regf=@(XTRAIN,ytrain,XTEST)(XTEST*regress(ytrain,XTRAIN));

cvMse = crossval('mse',X,y,'Predfun',regf,'Options',opts)

cvMse =

    0.1028

This simple example is not a good candidate for parallel computation:

% How long to compute in serial?
tic;cvMse = crossval('mse',X,y,'Predfun',regf);toc
Elapsed time is 0.073438 seconds.

% How long to compute in parallel?
tic;cvMse = crossval('mse',X,y,'Predfun',regf,...
    'Options',opts);toc
Elapsed time is 0.289585 seconds.

Reproducible Parallel Cross Validation

To run crossval in parallel in a reproducible fashion, set the options and reset the random stream appropriately (see Running Reproducible Parallel Computations).

mypool = parpool()

Starting parpool using the 'local' profile ... connected to 2 workers.

mypool = 

  Pool with properties:

    AttachedFiles: {0x1 cell}
       NumWorkers: 2
      IdleTimeout: 30
          Cluster: [1x1 parallel.cluster.Local]
     RequestQueue: [1x1 parallel.RequestQueue]
      SpmdEnabled: 1

s = RandStream('mlfg6331_64');
opts = statset('UseParallel',true,...
    'Streams',s,'UseSubstreams',true);

load('fisheriris');
y = meas(:,1);
X = [ones(size(y,1),1),meas(:,2:4)];
regf=@(XTRAIN,ytrain,XTEST)(XTEST*regress(ytrain,XTRAIN));

cvMse = crossval('mse',X,y,'Predfun',regf,'Options',opts)

cvMse =

    0.1020

Reset the stream:

reset(s)
cvMse = crossval('mse',X,y,'Predfun',regf,'Options',opts)

cvMse =

    0.1020