MATLAB: Convert a for loop to parfor loop

for looploop independentparfor

Hi,
In order to make my code faster, I am trying to convert for loop to parfor loop.
The original subpart of my code is (subject to error)
SNR_ind = -5:2.5:10;
run = 100;
count = 1;
for SNR_ind = SNR
    error_NP = [];
    error_SPVP = [];
    parfor i = 1:run
        %% Measurement
        [S_IC,S_IT,W_T,W_C] = noise_gen(SNR_ind,SIR,A_I,S_IC1,S_IT1,I,N_I,Avg_eeg_S.avg,t_index);
        [Avg_eeg_IC,~] = source_modeling(vol,elec,d_I_loc,t_index,S_IC); % With new S_IC,S_IT
        [Avg_eeg_IT,dip_posm] = source_modeling(vol,elec,d_I_loc,t_index,S_IT);
        X_T = Avg_eeg_S.avg + Avg_eeg_IT.avg + W_T;
        X_C = Avg_eeg_IC.avg + W_C;
        %% Inverse localization
        [error] = Null_proj(leadfield_eeg_scan,N_S,N_I,X_C,X_T,zplane,mri_resliced,d_S_loc);
        error_NP = [error_NP error];
        [error] = SPVP_proj(leadfield_eeg_scan,N_S,N_I,X_C,X_T,zplane,mri_resliced,d_S_loc);
        error_SPVP = [error_SPVP error];
    end
    Eerror_NP(count,:) = error_NP;
    Eerror_SPVP(count,:) = error_SPVP;
    count = count+1;
end
I have used parfor in inner loop as the inner loop will run more times than the outer loop.
I am getting error – " Error: The temporary variable 'S_IC' must be set before it is used. For more information, see Parallel for Loops in MATLAB, "Uninitialized Temporaries"."
I do understand the problem that variable S_IC need to be used in the very next statement it is calculated. But I don't know how to work around. I've read the documentation, but i can't figured out how to convert into a parfor loop. How should I use parfor in this context?

Best Answer

I figured out, how to solve serial for to parfor. The variable who are causing the dependency in the loop. I calculated them in a prior loop and called them afterwards. Also, it's better to use cells instead of arrays in context to parallel computing. Parfor decreased my computation time by 1/4. This is really nice to learn. The revised code of my earlier code is as follows.
SNR = -5:2.5:10;
run = 100;
count = 1;
for SNR_ind = SNR
    parfor i = 1:run
        [S_IC,S_IT,W_T,W_C] = noise_gen(SNR_ind,SIR,A_I,S_IC_initial,S_IT_initial,I,N_I,Avg_eeg_S.avg,t_index);
        S_IC1{1,i} = S_IC;
        S_IT1{1,i} = S_IT;
        W_T1{1,i} = W_T;
        W_C1{1,i} = W_C;
    end
    S_IC2{count,1} = S_IC1;
    S_IT2{count,1} = S_IT1;
    W_T2{count,1} = W_T1;
    W_C2{count,1} = W_C1;
    count = count+1;
end
count = 1;
for SNR_ind = SNR
    parfor i = 1:run
        [Avg_eeg_IC,~] = source_modeling(vol,elec,d_I_loc,t_index,S_IC2{count,1}{1,i}); % With new S_IC,S_IT

        Avg_eeg_IC1{1,i} = Avg_eeg_IC.avg;
        [Avg_eeg_IT,~] = source_modeling(vol,elec,d_I_loc,t_index,S_IT2{count,1}{1,i}); % With new S_IC,S_IT
        Avg_eeg_IT1{1,i} = Avg_eeg_IT.avg;
    end
    Avg_eeg_IC2{count,1} = Avg_eeg_IC1;
    Avg_eeg_IT2{count,1} = Avg_eeg_IT1;
    count = count+1;
end
count = 1;
for SNR_ind = SNR
    for i = 1:run
        X_T1{1,i} = Avg_eeg_S.avg + Avg_eeg_IT2{count,1}{1,i}+ W_T2{count,1}{1,i};
        X_C1{1,i} = Avg_eeg_IC2{count,1}{1,i}+ W_C2{count,1}{1,i};
    end
    X_T{count,1} = X_T1;
    X_C{count,1} = X_C1;
    count = count+1;
end
count = 1;
for SNR_ind = SNR
    parfor i = 1:run
        i
        close all
        %% Inverse localization
        [error_NP{1,i}] = Null_proj(leadfield_eeg_scan,N_S,N_I,X_C{count,1}{1,i},X_T{count,1}{1,i},zplane,mri_resliced,d_S_loc);
        [error_SPVP{1,i}] = SPVP_proj(leadfield_eeg_scan,N_S,N_I,X_C{count,1}{1,i},X_T{count,1}{1,i},zplane,mri_resliced,d_S_loc);
    end
    Eerror_NP{count,:} = cell2mat(error_NP);
    Eerror_SPVP{count,:} = cell2mat(error_SPVP);
    count = count+1;
end
Related Question