Trying to reduce computation time

Hey, I am trying to implement a code to compute the values of two functions using a large array of input.

visitst = symfun(exp(T/4)+1000,T);
visitsp = symfun(-10*(P-10)^2+5000,P);

for i=1:n
    Values1(i) = visitst(Input1(i));
    Values2(i) = visitsp(Input2(i));
end

I am looking to go through an "Input1" and "Input2" array of 100k+ different inputs, but even 1000 are taking extremely long computing times. Can someone suggest a method of improving this?

Answer 1

[Jan]

Pre-allocation the output before the loop:

Value1 = zeros(1, n);
Value2 = zeros(1, n);

I do not expect, that this is the bottleneck. But the costs of a forgotton pre-allocation grow exponentially, such that there is a limit in the input size, where this becomes the bottleneck.

[ThB]

That just cut the computational time massively (from 30sec to 4sec for example). I just preallocated the size using

Value1 = zeros(n);

Thank you very much!

Any other ideas on how to speed things up?

[Jan]

Wow, I'm surprised. Note that zeros(n) allocates a n*n matrix, but I'm not sure if the double class is sufficient for your case. Another method for an implicit pre-allocation is to run the loop backwards:

for i = n:-1:1  % Backwards for implicit pre-allocation

Then the last element is created at first, which reserves the complete vector at once - and in the matching class. Just be sure to add the comment, otherwise the readers (like you in 3 months) might wonder, what the purpose of this direction might be useful for.

For further speedups, use the profiler at first: Find the line, which uses the most processing time. Optimizing other parts is hardly useful.