All about the generic optimization framework available in scikits.openopt.solvers.optimizers
I’m please to announce a new version for scikits.optimization. The main focus of this iteration was to finish usual unconstrained optimization algorithms.
Changelog
The scikit can be installed with pip/easy_install or downloaded from PyPI
Old announces:
In the next version of scikits.optimization, I’ve added some Quasi-Newton steps. Before this version is released, I thought I would compare several methods of optimizing the Rosenbrock function.
Now that version 0.2 of scikit.optimization is out, here is a tutorial on the gradient-free optimizer based on the simplex algorithm.
When the only thing you have is the cost function and when you don’t have dozens of parameters, the first thing that can be tried is a simplex algorithm.
It has been a while, too long for sure, since my last update on this scikit. I’m pleased to announce that some algorithms are finally fixed as well as some tests.
Changelog:
Additional tutorials will be available in the next weeks.
Old announces:
I had to port a simplex/Nelder-Mead optimizer that I already have in Python in C++. As for the Python version, I tried to be as generic as possible but as efficient as possible, so the state is no longer a dictionary, but a simple structure.
I could have used the Numerical Recipes version, but the licence cost is not worth it, and the code is not generic enough, not explained enough. And also there are some design decisions that are questionable (one method = one responsibility).
My last blog post on optimization helped me generate orthogonal sequences. Now, I will use those sequences to separate two signals. The basic use case is a linear system with two inputs, one output, and instead of recording the response of one input at a time, one plays both inputs simultaneously with specific sequences so that they can be separated in another process.
Although I’m fond of numerical optimization through gradients, … there are some times where a global optimization is much more powerfull. For instance, I have to generate two sequences/combs that are orthogonal and for which their autocorrelation is almost an impulse. The two combs have a fixed number of impulse, so it’s a perfect job for genetic algorithms.
In my last post about optimization, I’ve derived my function analytically. Sometimes, it’s not as easy. Sometimes also, a simple gradient optimization is not enough.
scikits.optimization has a special class for handling numerical differentiation, and several tools for conjugate gradients.
Some weeks ago, the first release of the optimization scikit was done. I’d like to expose here the internal structure and the way the implementation was thought.