减少TSP动态编程实现的内存使用

Question

我正在尝试实施动态编程来解决一组25个城市的TSP问题。 通过直接实现程序，我面临“内存错误”。我试图通过考虑我的实现仅需要比当前子集小1的子集来解决它，即动态编程中的子问题是移除了一个元素的当前集合。因此，仅存储对应于当前和先前尺寸的距离（尺寸从2到25变化）。尽管如此，该程序仍占用了近85％的内存然后抛出错误。我的系统有4GB的RAM。 如何修复代码以使其内存使用不会爆炸？

def make_set(i):
    '''Generates all subsets of size i(integer)
        from cities, including the element 1'''
    cities = range(2,26)
    subs = comb(cities,i)
    for j in subs:
        j=set(j)
        j.add(1)
        yield j

def tsp(cities,dist):
    '''Performs search on subsets to find shortest path that traverses
       all cities starting and ending at 1
       Returns a dictionary, contains all 
       needs to be processed further to find the subset_dist answer'''
    subset_dist={}
    subset_dist[(frozenset([1]),1)]=0
    for i in range(1,25): 
        subset_dist_new={}
        subsets = make_set(i) #generator for subsets of size i
        for S in subsets:
            for j in S:       # j is the destination
                if j!=1:
                    min_dist= float('inf')
                    S_minusJ = set(S)
                    S_minusJ.remove(j)
                    for k in S:     #k defines the subproblem, the closest city before j, for current subset
                        if k==1 and len(S)>2:
                            continue
                        if k!=j:
                            min_dist = min(min_dist,subset_dist[(frozenset(S_minusJ),k)]+dist[frozenset([j,k])])
                            subset_dist_new[(frozenset(S),j)] = min_dist 
        subset_dist = subset_dist_new
    return subset_dist_new