Memoization

The nullspace_optimizer package includes the decorator memoize() for the memoization of functions. This allows to avoid multiple calls to an expensive function that needs to be used by several methods of an optimization problem, such as multiple constraints or the objective function and a constraint.

Usage

Suppose we have func is an expensive function that returns a dictionary res with the value res['J'] of the objective function and a single constraint res['G']. This function can be memoized as follows:

python

from nullspace_optimizer import memoize

# enable memoization
@memoize()
def func(x):
    # An expensive function
    # ....
    return res

Then a stored value will be returned if func is called several time with the same argument x. This can be useful if func is needed in multiple methods of an Optimizable object, such as J, G or H:

python

from nullspace_optimizer import Optimizable, nlspace_solve

class Problem(Optimizable):

    def x0(self):
    # Initialization...

    def J(self, x):
       # Some expensive computation
       res = func(x)
       return res['J']

    def G(self, x):
       # Some expensive computation
       res = func(x)
       return res['G']


 # Solve the optimization problem
 # Memoization will be used when calling J and G, func will be called
 # only once at every iteration.
 results = nlspace_solve(Problem())