Hi Maya,
As per my understanding the error message you received indicates that there is an attempt to access an index of an array that exceeds its bounds, which is a common issue when dealing with parallel loops due to the non-deterministic order of execution. Here are a few workarounds and suggestions to resolve the issue:
1. Preallocate and Manage Array Indexing:
Ensure each iteration of loop writes to a unique portion of array. One way to manage this is by preallocating a large enough array and calculating indices in a way that they do not overlap between iterations. An alternative approach is to use cell arrays or structures where one writes to a separate cell or field, which are then concatenated or merged after the loop.
2. Use parfor with Independent Iterations:
Ensure iterations of “parfor” loop are independent. Error could be due to dependencies between iterations that are not apparent. Given the current indexing strategy is dependent on the cumulative sum of m, consider redesigning this part. A solution approach-
- Independent Data Collection: Use “parfor” to collect data in a way that each iteration's output is independent and does not rely on previous iterations. This might involve using a cell array or struct array to store results.
- Post-Processing: After collecting all data independently, perform any necessary sequential operations on the data outside of the “parfor” loop.
Here is a conceptual example to adjust the code:
% Assuming LQ_LQe_Simu returns vectors/arrays that can vary in size,
% Use a cell array to collect outputs
numIterations = ...; % Calculate the total number of iterations
results = cell(numIterations, 1);
parfor idx = 1:numIterations
[sigma, alpha, beta] = ...; % Calculate or retrieve sigma, alpha, beta for this iteration
[indices, observation] = LQ_LQe_Simu(sigma, alpha, beta);
% Store in a cell array or struct
results{idx} = struct('indices', indices, 'observation', observation, 'params', [sigma, alpha, beta]);
end
% Convert results from cell array to a structured format that you need
mat_1 = ...; % Initialize your matrix or other data structure
for idx = 1:numIterations
% Extract data from results{idx} and populate mat_1 or other variables
% This is where you would handle the logic currently inside your nested loops
End
Please refer to following documentation links-
- Parfor- https://www.mathworks.com/help/matlab/ref/parfor.html?searchHighlight=Parfor&s_tid=srchtitle_support_results_1_Parfor
- Cell- https://www.mathworks.com/help/matlab/ref/cell.html?searchHighlight=cell&s_tid=srchtitle_support_results_1_cell
- Struct- https://www.mathworks.com/help/matlab/ref/struct.html?searchHighlight=struct&s_tid=srchtitle_support_results_1_struct
3. Debugging and Validation:
Before running the full “parfor” loop, validate the approach with small number of iterations to ensure correctness. Ensure that any dynamic calculation of indices is correctly implemented, especially under the parallel execution model where iterations run in an unpredictable order.
Transitioning from serial to parallel computation, especially in complex data processing scenarios, requires rethinking the data flow and processing strategy to fit the parallel model. Ensuring data independence between iterations and correctly managing data aggregation post-parallel processing are key to leveraging “parfor” effectively.
Hope it helps!
Best Regards,
Simar
