Hello everyone
I receive the following type of warning message when I use the Matlab local search function ‘fmincon’:
Warning: Matrix is close to singular or badly scaled. Results may be inaccurate. RCOND = 1.343564e-18.
So, does this mean my results will be influenced?
PS: I am working on CEC'14 benchmarks; I need the maximum of accuracy of results.
Any help would be really appreciable.
This is a general problem in matrix algebra when the mathematical solution is based on inverting a matrix. Singular matrices do not have an inverse. The application is telling you that your data are not suitable for the specific analysis as is. Some transformation or scaling could help depending on the nature of the data. It might also mean that the equations being used for the analysis are inappropriate (difficult to tell).
Yes, or at least maybe, with respect to will your results be influenced. The further RCOND (reciprocal condition number) is from 1 the less well conditioned the data in the context of the analysis.
I agree that it's hard to say much without runnable code.
Are you supplying your own gradient or Hessian to fmincon? If yes, the be sure to use the 'DerivativeCheck' option via optimoptions and compare the result to the default finite-differencing schemes to make sure you don't have a bug:
http://mathworks.com/help/optim/ug/fmincon.html#f605851
You should also check your objective function for bugs and look for any ways to improve it numerically (e.g., re-scaling the problem to avoid small or large values and avoiding under- and overflow situations).
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence