improve for loop performance

Hi,

I have a for loop runing on all values and takes into acount previous value

for i=2:length(a)
    a(i)=some_function(input1, input2, a(i-1))
end

where a is of size 200,000-300,000

Is there a way to improve it and speed up matlab performance?

Answer 1

[Walter Roberson]

In general:

• you could get higher performance by moving the loop inside of some_function, so as to reduce the number of function calls
• otherwise, it depends what is being calculated inside the function. The kind of code you are using is similar to what would be used to "filter" a signal, and if you can rewrite to use filt() or filtfilt() then that might be faster -- if the processing can be written in terms of derivatives or in terms of examining previous time steps. Also, if the computation is in terms of derivatives or timesteps, then Control System Toolbox transfer functions or state space models are sometimes good representations.
• But there are certainly some computations that filter() and transfer functions are irrelevant for. There just might not be any useful speed improvement (other than running the calculation inside of one function to save on the function calls.)

[michael]

The some_function is used to calculate some value including the hysteresis around it.

If a(i) is out from hysteresis then select new valaue, else no change from previous value.

[Walter Roberson]

It sounds like input1 and input2 are not constants then?

[michael]

Right They are the input values for current point for the hysteresis (x,y)