spmd running multiple times

I'm running 2 RBF networks with different spread constant values. I'm trying to use spmd to get both executed in concurrent fashion. Following is the code:

 spread1=365;
 spread2=575;
 goal=0.0;
 max_neurons=2500;
 neuron_jump=500;
 tic
   spmd(2)
     RB_SC_365=newrb(input_mat, targets, goal, spread1, max_neurons, neuron_jump);
     RB_SC_575=newrb(input_mat, targets, goal, spread2, max_neurons, neuron_jump);
   end
 toc

Per my understanding of spmd, spmd(2) will take 2 worker threads from the pool, one will execute RB_SC_365 and other will execute RB_SC_575, and then the script will exit. However, the script is being executed twice.

Here is the output I'm getting:

Lab 1: 
  NEWRB, neurons = 0, MSE = 0.16
Lab 2: 
  NEWRB, neurons = 0, MSE = 0.16
  NEWRB, neurons = 500, MSE = 1.14309e-28
Lab 1: 
  NEWRB, neurons = 500, MSE = 1.14309e-28
Lab 2: 
  NEWRB, neurons = 0, MSE = 0.16
Lab 1: 
  NEWRB, neurons = 0, MSE = 0.16
Lab 2: 
  NEWRB, neurons = 500, MSE = 5.26541e-29
Lab 1: 
  NEWRB, neurons = 500, MSE = 5.26541e-29

As can be seen above, the script has executed twice. The output that I expect is:

Lab 1: 
  NEWRB, neurons = 0, MSE = 0.16
Lab 2: 
  NEWRB, neurons = 0, MSE = 0.16
  NEWRB, neurons = 500, MSE = 1.14309e-28
Lab 1: 
  NEWRB, neurons = 500, MSE = 1.14309e-28

Is there some way to execute the script only once with spmd?

Answer 1

[Walter Roberson]

You wrote,

         spmd(2)
     RB_SC_365=newrb(input_mat, targets, goal, spread1, max_neurons, neuron_jump);
     RB_SC_575=newrb(input_mat, targets, goal, spread2, max_neurons, neuron_jump);
   end

"Per my understanding of spmd, spmd(2) will take 2 worker threads from the pool, one will execute RB_SC_365 and other will execute RB_SC_575"

No, that is not correct. All workers will execute what you have inside the spmd directive. You would need to do

         spmd(2)
           if labindex == 1
     RB_SC_365=newrb(input_mat, targets, goal, spread1, max_neurons, neuron_jump);
           else
     RB_SC_575=newrb(input_mat, targets, goal, spread2, max_neurons, neuron_jump);
           end
   end

[Akshay Joshi]

Hi Walter,

Thanks a lot. I've tried this solution on a small dataset, and it's working perfectly.

Just generalizing the question, if we want to run more than 2 statements, then

spmd(2)
           if labindex == 1
     RB_SC_365=newrb(input_mat, targets, goal, spread1, max_neurons, neuron_jump);
           else if labindex == 2
     RB_SC_575=newrb(input_mat, targets, goal, spread2, max_neurons, neuron_jump);
else
%%some other statement
           end
   end

Is my understanding correct? Please advise.

[Walter Roberson]

Some people find it cleaner to use switch and case instead of if/elseif, but the functionality is the same.

With spmd(2) you will only ever have two labs, so labindex == 1 and labindex == 2 are the only possibilities; your second else will not be reached.

To emphasize:

for spmd, any statement that is not protected by a test against labindex will be executed by all of the labs. But that does not necessarily mean that they will all be working on the same data. For example,

spmd(10)
  filename_in = sprintf('black_mantis_%03d.png', labindex);
  filename_out = sprintf('blacker_mantis_%03d.png', labindex);
  img = im2double(rgb2gray(imread(filename_in)));
  new_img = img.^2;
  imwrite(new_img, filename_out);
end

The same code is executed for all of the labs, but working with different files.