[SciPy-user] Linear Constrained Quadratic Optimization Problem
Tue Aug 28 13:29:22 CDT 2007
afaik scipy hasn't an appropriate solver.
The only one that handles other than lb-ub constraints is
scipy.optimize.fmin_cobyla, but you should type each constraint as
separate one, i.e. Ax<=b or Aeq x = beq can't be pass as matrices. Also,
cobyla can't handle user-supplied gradients.
You can try scikits.openopt, solver lincher. Not far from best for now,
but, at least, it's better than nothing.
svn co http://svn.scipy.org/svn/scikits/trunk/openopt openopt
sudo python setup.py install
from scikits.openopt import NLP
you should use lincher solver, also, you can assign penalties to your
linear constraints and use unconstrained ralg solver, it's capable of
handling rather huge penalties.
I hope I will enhance the solver lincher from time to time. Also, I hope
soon ALGENCAN will be connected, and/or NLPy, when it will finish
migrating from numeric to numpy
Also, I hope some months or weeks later ralg-based QP/QPQC solver will
be ready, it can solve your problem even with quadratic constraints.
Chris Hudzik wrote:
> I am trying to solve a quadratic optimization problem:
> minimize (x - y*p)^T * V * (x - y*p) over x,y
> x^T * t = b
> where x, t, and p are vectors; V is a symmetric matrix; and y and b are scalars.
> I am new to scipy. Can anybody point me to a scipy module to solve this?
> SciPy-user mailing list
More information about the SciPy-user