There are two mainstream optimization packages available:
OpenOpt (license: BSD)
SciPy.optimize (license: BSD)
scipy.optimize contains only copyleft-free solvers, while OpenOpt contains connections to any-licensed solvers (BSD, GPL, LGPL etc), as well as some our own Python-written BSD-licensed solvers (numpy required), has graphic output of convergence and some more numerical optimization "MUST HAVE" features. Also OpenOpt can solve FuncDesigner problems with automatic differentiation, that usually work faster and gives more precize results than finite-differences derivatives approximation.