parfor error for global SA Morris method using SimBiology

Question

emjey 2019-8-26

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/477595-parfor-error-for-global-sa-morris-method-using-simbiology

评论： emjey 2019-8-26

hi, I am trying to use parallel toolbox to speed up my global SA with Morris method. Basically it averages elementary effects for a specific parameters across the entire parameter space. An elementary effect is calculated as EE = (f(p) - f(p+delta))/delta. For more details see Wentworth et al. J. UNCERTAINTY QUANTIFICATION, Vol. 4, pp. 266–297 (https://projects.ncsu.edu/crsc/reports/ftp/pdf/crsc-tr15-01.pdf).

So far I failed with my attempts, such as for this code - here using SimBiology 'sbioselect' and 'sbiosimulate'

clear variables; close all; clc;
format shortG
r = 5; % number of trajectories
l = 10; s = 2;
delta = l/(s*(l-1));
disp("Load model & parameter file")
paramsFile = strcat('PARAMETERS/Lotka_parameters.csv');
paramsDF = readtable(paramsFile);
sbioloadproject('SBPROJ/lotka.sbproj','m1')
p = length(paramsDF.parameter);
% Morris ===========================================================
d = zeros(r,p);
for i = 1:r
    
    f = zeros(1,(p+1));
    % an example for the Morris C matrix capturing one trajectory in parameter space
    C = [15     0.085556           15;
        15        0.005           15;
        15        0.005       9.4444;
        9.4444        0.005       9.4444];
    
    % for each trajectory for p parameters, there are p+1 evaluations to be made
    for j = 1:(p+1)                              
        for k = 1:size(C,2) % columns = parameters
            blub = sbioselect(m1, 'Name', paramsDF.parameter(k));
            blub.Value = C(j,k);
        end
        
        [t,y] = sbiosimulate(m1);
        f(j) = y(end,varNo);      % looking for the endpoint as the sensitivity function
    end
    
    % Calculating elementary effects
    v = zeros(1,p);
    for j = 1:p
        Cindex = find(diff(C(:,j)) ~= 0);
        v(j) = ((f(Cindex) - f(Cindex+1))/delta;
    end
    d(i,:) = v;
end
%% disp("Calculating EE statistics")
... based on d(i,j)

Although the code is correct from the syntaxt point of view, at run time I get the error:

Load model & parameter file
Starting parallel pool (parpool) using the 'local' profile ...
connected to 4 workers.
Error using sbiosimulate (line 136)
Expected input number 1, MOBJ, to be one of these types:
SimBiology.Model
Instead its type was double.
Error in test (line 19)
parfor i = 1:r

Any comments would be very appreciated. I attach lotka.sbproj and parametr file as well.

Best, M

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Florian Augustin 2019-8-26

3
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/477595-parfor-error-for-global-sa-morris-method-using-simbiology#answer_389128

Hi,

It looks like you are running into an issue that the SimBiology model is not copied to the workers natively. In MATLAB versions R2019a and earlier, you have to use parallel.pool.Constant to send the model to the workers in your parallel pool. The following tweak to your code should do this for you.

m1ParConst = parallel.pool.Constant(m1);
parfor i = 1:r
    % ...
        % Get the model from the parPool constant as follows:
        blub = sbioselect(m1ParConst.Value, 'Name', paramsDF.parameter(k));
    % ...
end