[SciPy-user] Simulated annealing in scipy

joris at ster.kuleuven.ac.be joris at ster.kuleuven.ac.be
Fri Jul 22 16:35:42 CDT 2005


Hi,

An interesting alternative to GAs or SA for global optimization is the
so-called Neighbourhood-algorithm:
http://wwwrses.anu.edu.au/~malcolm/na/na.html

Features of this algorithm are:
- suitable for a multidimensional parameter space
- derivative free
- it only uses the rank of the function to be optimized
  (that is, no need to rescale your function to make it work)
- only 2 control parameters
- a theoretical framework to derive estimates of your model parameters, 
  the covariance matrix, and even Bayesian posteriori PDFs, has been 
  developed and implemented
- well-documented
- Fortran code is freely available upon request

I've used this algorithm in the past, with success.

Cheers,
Joris






More information about the SciPy-user mailing list