In any pattern recognition model complexity is one of the important factors that needs to be taken into account. Using Matlab, what are the possible ways to measure this factor? The tic toc time computing is well known for measuring the consumed time for executing any code, but can anyone suggest any other way for the same purpose?
Is this two questions? Or does the question concern the code complexity of the model implementation? "mlint" and "checkcode" will calculate cyclomatic complexity, in matlab. "profile" can help develop call graph and identify high overhead bits.
Complexity for machine learning problems, though, usually concerns complexity of the classification problem in the feature space. Many complexity measures exist, separability, number of unique class clusters, entropy and information measures, for example. Mathworks has the statistics, neural network and fuzzy logic toolboxes; CLOP and spider are also freeware toolboxes for matlab, .
Several other academic toolboxes are out there, I see http://staffwww.dcs.shef.ac.uk/people/N.Lawrence/mltools/ for instance - your mileage may vary.
Thanks James, I have two different way feature preprocessing which feed SVM classifier. I want to know, which one of these two ways is more complex. I tried to calculate the time of implementing the code of both of the two ways. Perhaps I have to ask how measuring the code computation time is a proper indication of the model complexity?
- Badges
- Science topic