I'm working on a simple resource allocation problem, that I'm solving using backward DP. The full code is at: https://codereview.stackexchange.com/questions/123641/allocating-a-resource

It works fine, but I'm a bit puzzled on how to retrieve the optimal policy once the optimisation is done. I use a memoization decorator but I'm clueless as to how to get the optimal policy out of it at the end. (The policy would be a series of 1,0 as describing when to allocate and when not to.)

def memoize(f):
    cache = {}

    def memoizedFunction(*args):
        if args not in cache:
            cache[args] = f(*args)
        return cache[args]

    memoizedFunction.cache = cache
    return memoizedFunction

In an alternative effort I use:

def record(D,key,item):
    """Records the value of item in the dictionary d with k"""
    if key in D.keys():
        D[key] = [item]

    return D

And in the DP:

   if v == no_pump:
        pos = 0
        pos = 1


My hunch is that if I can create a new dictionary (X) for each policy, and record the optimum value up to its execution, it could work. But I don't know how to do that. Also it seams quite inefficient.

To clarify: Forwards methods, or converting it into a graph aren't suitable as the whole problem will be converted into a stochastic control problem later.


Right now the function you memoize simply returns the score the best solution. Change is so that it returns a tuple of the first step of the best solution and the score. Something like:

def score(...)
    _, zero_score = score(...)
    _, one_score = score(...)
    if zero_score < one_score:
       return 0, zero_score
       return 1, one_score

Now, to find out the next step, you simply have to call score and it will tell you the next step and the score.


Move your memoizedFunction.cache dictionary out of the function memoize function so it is at a global level. Then you will be able to query it after the optimization has run.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.