一组联合查找算法

时间:2010-06-18 06:05:34

标签: python algorithm set

我有数千行1到100个数字,每行定义一组数字以及它们之间的关系。 我需要得到一组相关的数字。

小例子: 如果我有这7行数据

T1 T2
T3 
T4
T5
T6 T1
T5 T4
T3 T4 T7

我需要一个不那么慢的算法来知道这里的集合是:

T1 T2 T6 (because T1 is related with T2 in the first line and T1 related with T6 in the line 5)
T3 T4 T5 T7 (because T5 is with T4 in line 6 and T3 is with T4 and T7 in line 7)

但是当你拥有非常大的集合时,在每个大集合中搜索T(x)都会非常缓慢,并且需要集合等等。

你是否有提示以不那么强力的方式做到这一点?

我正在尝试用Python做这件事。

5 个答案:

答案 0 :(得分:14)

将您的数字T1,T2等视为图顶点。在一条线上出现的任何两个数字由边连接。那么您的问题就等于找到此图表中的所有connected components。您可以通过从T1开始,然后执行广度优先或深度优先搜索来查找从该点可到达的所有顶点。 将所有这些顶点标记为属于等价类T1。然后找到下一个未标记的顶点Ti,找到从那里可到达的所有尚未标记的节点,并将它们标记为属于等价类Ti。继续,直到标记了所有顶点。

对于具有n个顶点和e边的图,该算法需要O(e)时间和空间来构建邻接列表,并且O(n)时间和空间用于标识所有连接的组件 一旦构建了图形结构。

答案 1 :(得分:14)

一旦构建了数据结构,您想要针对它运行哪些查询?向我们展示您现有的代码。什么是T(x)?你谈到“数字组”,但你的样本数据显示T1,T2等;请解释一下。

你读过这个:http://en.wikipedia.org/wiki/Disjoint-set_data_structure

尝试查看此Python实现:http://code.activestate.com/recipes/215912-union-find-data-structure/

或者你可以自己抨击一些相当简单易懂的东西,例如

[更新:全新代码]

class DisjointSet(object):

    def __init__(self):
        self.leader = {} # maps a member to the group's leader
        self.group = {} # maps a group leader to the group (which is a set)

    def add(self, a, b):
        leadera = self.leader.get(a)
        leaderb = self.leader.get(b)
        if leadera is not None:
            if leaderb is not None:
                if leadera == leaderb: return # nothing to do
                groupa = self.group[leadera]
                groupb = self.group[leaderb]
                if len(groupa) < len(groupb):
                    a, leadera, groupa, b, leaderb, groupb = b, leaderb, groupb, a, leadera, groupa
                groupa |= groupb
                del self.group[leaderb]
                for k in groupb:
                    self.leader[k] = leadera
            else:
                self.group[leadera].add(b)
                self.leader[b] = leadera
        else:
            if leaderb is not None:
                self.group[leaderb].add(a)
                self.leader[a] = leaderb
            else:
                self.leader[a] = self.leader[b] = a
                self.group[a] = set([a, b])

data = """T1 T2
T3 T4
T5 T1
T3 T6
T7 T8
T3 T7
T9 TA
T1 T9"""
# data is chosen to demonstrate each of 5 paths in the code
from pprint import pprint as pp
ds = DisjointSet()
for line in data.splitlines():
    x, y = line.split()
    ds.add(x, y)
    print
    print x, y
    pp(ds.leader)
    pp(ds.group)

这是最后一步的输出:

T1 T9
{'T1': 'T1',
 'T2': 'T1',
 'T3': 'T3',
 'T4': 'T3',
 'T5': 'T1',
 'T6': 'T3',
 'T7': 'T3',
 'T8': 'T3',
 'T9': 'T1',
 'TA': 'T1'}
{'T1': set(['T1', 'T2', 'T5', 'T9', 'TA']),
 'T3': set(['T3', 'T4', 'T6', 'T7', 'T8'])}

答案 2 :(得分:2)

您可以使用联合查找数据结构来实现此目标。

这种算法的伪代码如下:

func find( var element )
    while ( element is not the root ) element = element's parent
    return element
end func

func union( var setA, var setB )
    var rootA = find( setA ), rootB = find( setB )
    if ( rootA is equal to rootB ) return
    else
        set rootB as rootA's parent
end func

(取自http://www.algorithmist.com/index.php/Union_Find

答案 3 :(得分:1)

正如上面提到的Jim,您实际上是在寻找简单无向图的connected components,其中节点是您的实体(T1T2等等) ,edge表示它们之间的成对关系。连接组件搜索的简单实现基于广度优先搜索:从第一个实体启动BFS,找到所有相关实体,然后从第一个尚未发现的实体启动另一个BFS,依此类推,直到找到它们为止所有。 BFS的简单实现如下:

class BreadthFirstSearch(object):
    """Breadth-first search implementation using an adjacency list"""

    def __init__(self, adj_list):
        self.adj_list = adj_list

    def run(self, start_vertex):
        """Runs a breadth-first search from the given start vertex and
        yields the visited vertices one by one."""
        queue = deque([start_vertex])
        visited = set([start_vertex])
        adj_list = self.adj_list

        while queue:
            vertex = queue.popleft()
            yield vertex
            unseen_neis = adj_list[vertex]-visited
            visited.update(unseen_neis)
            queue.extend(unseen_neis)

def connected_components(graph):
    seen_vertices = set()
    bfs = BreadthFirstSearch(graph)
    for start_vertex in graph:
        if start_vertex in seen_vertices:
            continue
        component = list(bfs.run(start_vertex))
        yield component
        seen_vertices.update(component)

这里,adj_listgraph是一个邻接列表数据结构,基本上它给出了图中给定顶点的邻居。要从您的文件构建它,您可以这样做:

adj_list = defaultdict(set)
for line in open("your_file.txt"):
    parts = line.strip().split()
    v1 = parts.pop(0)
    adj_list[v1].update(parts)
    for v2 in parts:
        adj_list[v2].add(v1)

然后你可以运行:

components = list(connected_components(adj_list))

当然,在纯Python中实现整个算法往往比使用更高效的图形数据结构的C中的实现慢。您可以考虑使用igraph或其他图表库(如NetworkX)来完成工作。两个库都包含连接组件搜索的实现;在igraph中,它归结为此(假设您的文件不包含具有单个条目的行,仅接受成对条目):

>>> from igraph import load
>>> graph = load("edge_list.txt", format="ncol", directed=False)
>>> components = graph.clusters()
>>> print graph.vs[components[0]]["name"]
['T1', 'T2', 'T6']
>>> print graph.vs[components[1]]["name"]
['T3', 'T4', 'T5']

免责声明:我是igraph

的作者之一

答案 4 :(得分:0)

您可以使用set为群组建模。在下面的示例中,我将集合放入Group类中,以便更容易保持对它们的引用并跟踪一些名义上的“head”项。

class Group:
    def __init__(self,head):
        self.members = set()
        self.head = head
        self.add(head)
    def add(self,member):
        self.members.add(member)
    def union(self,other):
        self.members = other.members.union(self.members)

groups = {}

for line in open("sets.dat"):
    line = line.split()
    if len(line) == 0:
        break
    # find the group of the first item on the row
    head = line[0]
    if head not in groups:
        group = Group(head)
        groups[head] = group
    else:
        group = groups[head]
    # for each other item on the row, merge the groups
    for node in line[1:]:
        if node not in groups:
            # its a new node, straight into the group
            group.add(node)
            groups[node] = group
        elif head not in groups[node].members:
            # merge two groups
            new_members = groups[node]
            group.union(new_members)
            for migrate in new_members.members:
                groups[migrate] = group
# list them
for k,v in groups.iteritems():
    if k == v.head:
        print v.members

输出是:

set(['T6', 'T2', 'T1'])
set(['T4', 'T5', 'T3'])