完整图表的效率问题

Question

我试图实现一个函数来生成一个完整的图形。功能还可以，但是当我尝试用28k节点运行它时，我的笔记本电脑崩溃了。我的实施有问题吗？

def make_complete_graph(num_nodes):
    """
     Takes the number of nodes num_nodes and returns a dictionary corresponding
     to a complete directed graph with the specified number of nodes. A 
     complete graph contains all possible edges subject to the restriction 
     that self-loops are not allowed. The nodes of the graph should be numbered
     0 to num_nodes - 1 when num_nodes is positive. Otherwise, the function
     returns a dictionary corresponding to the empty graph.
    """

    if num_nodes<=0:
        return dict()
    else:
        return {key:set(range(num_nodes))-set([key]) for key in range(num_nodes)}

我使用在anaconda下运行的Spyder，我的操作系统是ubuntu 15.04

编辑：我做了改进但仍然崩溃

def make_complete_graph(num_nodes):
    """
     Takes the number of nodes num_nodes and returns a dictionary corresponding
     to a complete directed graph with the specified number of nodes. A 
     complete graph contains all possible edges subject to the restriction 
     that self-loops are not allowed. The nodes of the graph should be numbered
     0 to num_nodes - 1 when num_nodes is positive. Otherwise, the function
     returns a dictionary corresponding to the empty graph.
    """
    nodes = set(range(num_nodes))
    if num_nodes<=0:
        return dict()
    else:
        return {key:nodes.difference(set([key])) for key in nodes}