金狮镖局 Design By www.egabc.com

本文实例讲述了Python图算法。分享给大家供大家参考,具体如下:

#encoding=utf-8
import networkx,heapq,sys
from matplotlib import pyplot
from collections import defaultdict,OrderedDict
from numpy import array
# Data in graphdata.txt:
# a b  4
# a h  8
# b c  8
# b h  11
# h i  7
# h g  1
# g i  6
# g f  2
# c f  4
# c i  2
# c d  7
# d f  14
# d e  9
# f e  10
def Edge(): return defaultdict(Edge)
class Graph:
  def __init__(self):
    self.Link = Edge()
    self.FileName = ''
    self.Separator = ''
  def MakeLink(self,filename,separator):
    self.FileName = filename
    self.Separator = separator
    graphfile = open(filename,'r')
    for line in graphfile:
      items = line.split(separator)
      self.Link[items[0]][items[1]] = int(items[2])
      self.Link[items[1]][items[0]] = int(items[2])
    graphfile.close()
  def LocalClusteringCoefficient(self,node):
    neighbors = self.Link[node]
    if len(neighbors) <= 1: return 0
    links = 0
    for j in neighbors:
      for k in neighbors:
        if j in self.Link[k]:
          links += 0.5
    return 2.0*links/(len(neighbors)*(len(neighbors)-1))
  def AverageClusteringCoefficient(self):
    total = 0.0
    for node in self.Link.keys():
      total += self.LocalClusteringCoefficient(node)
    return total/len(self.Link.keys())
  def DeepFirstSearch(self,start):
    visitedNodes = []
    todoList = [start]
    while todoList:
      visit = todoList.pop(0)
      if visit not in visitedNodes:
        visitedNodes.append(visit)
        todoList = self.Link[visit].keys() + todoList
    return visitedNodes
  def BreadthFirstSearch(self,start):
    visitedNodes = []
    todoList = [start]
    while todoList:
      visit = todoList.pop(0)
      if visit not in visitedNodes:
        visitedNodes.append(visit)
        todoList = todoList + self.Link[visit].keys()
    return visitedNodes
  def ListAllComponent(self):
    allComponent = []
    visited = {}
    for node in self.Link.iterkeys():
      if node not in visited:
        oneComponent = self.MakeComponent(node,visited)
        allComponent.append(oneComponent)
    return allComponent
  def CheckConnection(self,node1,node2):
    return True if node2 in self.MakeComponent(node1,{}) else False
  def MakeComponent(self,node,visited):
    visited[node] = True
    component = [node]
    for neighbor in self.Link[node]:
      if neighbor not in visited:
        component += self.MakeComponent(neighbor,visited)
    return component
  def MinimumSpanningTree_Kruskal(self,start):
    graphEdges = [line.strip('\n').split(self.Separator) for line in open(self.FileName,'r')]
    nodeSet = {}
    for idx,node in enumerate(self.MakeComponent(start,{})):
      nodeSet[node] = idx
    edgeNumber = 0; totalEdgeNumber = len(nodeSet)-1
    for oneEdge in sorted(graphEdges,key=lambda x:int(x[2]),reverse=False):
      if edgeNumber == totalEdgeNumber: break
      nodeA,nodeB,cost = oneEdge
      if nodeA in nodeSet and nodeSet[nodeA] != nodeSet[nodeB]:
        nodeBSet = nodeSet[nodeB]
        for node in nodeSet.keys():
          if nodeSet[node] == nodeBSet:
            nodeSet[node] = nodeSet[nodeA]
        print nodeA,nodeB,cost
        edgeNumber += 1
  def MinimumSpanningTree_Prim(self,start):
    expandNode = set(self.MakeComponent(start,{}))
    distFromTreeSoFar = {}.fromkeys(expandNode,sys.maxint); distFromTreeSoFar[start] = 0
    linkToNode = {}.fromkeys(expandNode,'');linkToNode[start] = start
    while expandNode:
      # Find the closest dist node
      closestNode = ''; shortestdistance = sys.maxint;
      for node,dist in distFromTreeSoFar.iteritems():
        if node in expandNode and dist < shortestdistance:
          closestNode,shortestdistance = node,dist
      expandNode.remove(closestNode)
      print linkToNode[closestNode],closestNode,shortestdistance
      for neighbor in self.Link[closestNode].iterkeys():
        recomputedist = self.Link[closestNode][neighbor]
        if recomputedist < distFromTreeSoFar[neighbor]:
          distFromTreeSoFar[neighbor] = recomputedist
          linkToNode[neighbor] = closestNode
  def ShortestPathOne2One(self,start,end):
    pathFromStart = {}
    pathFromStart[start] = [start]
    todoList = [start]
    while todoList:
      current = todoList.pop(0)
      for neighbor in self.Link[current]:
        if neighbor not in pathFromStart:
          pathFromStart[neighbor] = pathFromStart[current] + [neighbor]
          if neighbor == end:
            return pathFromStart[end]
          todoList.append(neighbor)
    return []
  def Centrality(self,node):
    path2All = self.ShortestPathOne2All(node)
    # The average of the distances of all the reachable nodes
    return float(sum([len(path)-1 for path in path2All.itervalues()]))/len(path2All)
  def SingleSourceShortestPath_Dijkstra(self,start):
    expandNode = set(self.MakeComponent(start,{}))
    distFromSourceSoFar = {}.fromkeys(expandNode,sys.maxint); distFromSourceSoFar[start] = 0
    while expandNode:
      # Find the closest dist node
      closestNode = ''; shortestdistance = sys.maxint;
      for node,dist in distFromSourceSoFar.iteritems():
        if node in expandNode and dist < shortestdistance:
          closestNode,shortestdistance = node,dist
      expandNode.remove(closestNode)
      for neighbor in self.Link[closestNode].iterkeys():
        recomputedist = distFromSourceSoFar[closestNode] + self.Link[closestNode][neighbor]
        if recomputedist < distFromSourceSoFar[neighbor]:
          distFromSourceSoFar[neighbor] = recomputedist
    for node in distFromSourceSoFar:
      print start,node,distFromSourceSoFar[node]
  def AllpairsShortestPaths_MatrixMultiplication(self,start):
    nodeIdx = {}; idxNode = {}; 
    for idx,node in enumerate(self.MakeComponent(start,{})):
      nodeIdx[node] = idx; idxNode[idx] = node
    matrixSize = len(nodeIdx)
    MaxInt = 1000
    nodeMatrix = array([[MaxInt]*matrixSize]*matrixSize)
    for node in nodeIdx.iterkeys():
      nodeMatrix[nodeIdx[node]][nodeIdx[node]] = 0
    for line in open(self.FileName,'r'):
      nodeA,nodeB,cost = line.strip('\n').split(self.Separator)
      if nodeA in nodeIdx:
        nodeMatrix[nodeIdx[nodeA]][nodeIdx[nodeB]] = int(cost)
        nodeMatrix[nodeIdx[nodeB]][nodeIdx[nodeA]] = int(cost)
    result = array([[0]*matrixSize]*matrixSize)
    for i in xrange(matrixSize):
      for j in xrange(matrixSize):
        result[i][j] = nodeMatrix[i][j]
    for itertime in xrange(2,matrixSize):
      for i in xrange(matrixSize):
        for j in xrange(matrixSize):
          if i==j:
            result[i][j] = 0
            continue
          result[i][j] = MaxInt
          for k in xrange(matrixSize):
            result[i][j] = min(result[i][j],result[i][k]+nodeMatrix[k][j])
    for i in xrange(matrixSize):
      for j in xrange(matrixSize):
        if result[i][j] != MaxInt:
          print idxNode[i],idxNode[j],result[i][j]
  def ShortestPathOne2All(self,start):
    pathFromStart = {}
    pathFromStart[start] = [start]
    todoList = [start]
    while todoList:
      current = todoList.pop(0)
      for neighbor in self.Link[current]:
        if neighbor not in pathFromStart:
          pathFromStart[neighbor] = pathFromStart[current] + [neighbor]
          todoList.append(neighbor)
    return pathFromStart
  def NDegreeNode(self,start,n):
    pathFromStart = {}
    pathFromStart[start] = [start]
    pathLenFromStart = {}
    pathLenFromStart[start] = 0
    todoList = [start]
    while todoList:
      current = todoList.pop(0)
      for neighbor in self.Link[current]:
        if neighbor not in pathFromStart:
          pathFromStart[neighbor] = pathFromStart[current] + [neighbor]
          pathLenFromStart[neighbor] = pathLenFromStart[current] + 1
          if pathLenFromStart[neighbor] <= n+1:
            todoList.append(neighbor)
    for node in pathFromStart.keys():
      if len(pathFromStart[node]) != n+1:
        del pathFromStart[node]
    return pathFromStart
  def Draw(self):
    G = networkx.Graph()
    nodes = self.Link.keys()
    edges = [(node,neighbor) for node in nodes for neighbor in self.Link[node]]
    G.add_edges_from(edges)
    networkx.draw(G)
    pyplot.show()
if __name__=='__main__':
  separator = '\t'
  filename = 'C:\\Users\\Administrator\\Desktop\\graphdata.txt'
  resultfilename = 'C:\\Users\\Administrator\\Desktop\\result.txt'
  myGraph = Graph()
  myGraph.MakeLink(filename,separator)
  print 'LocalClusteringCoefficient',myGraph.LocalClusteringCoefficient('a')
  print 'AverageClusteringCoefficient',myGraph.AverageClusteringCoefficient()
  print 'DeepFirstSearch',myGraph.DeepFirstSearch('a')
  print 'BreadthFirstSearch',myGraph.BreadthFirstSearch('a')
  print 'ShortestPathOne2One',myGraph.ShortestPathOne2One('a','d')
  print 'ShortestPathOne2All',myGraph.ShortestPathOne2All('a')
  print 'NDegreeNode',myGraph.NDegreeNode('a',3).keys()
  print 'ListAllComponent',myGraph.ListAllComponent()
  print 'CheckConnection',myGraph.CheckConnection('a','f')
  print 'Centrality',myGraph.Centrality('c')
  myGraph.MinimumSpanningTree_Kruskal('a')
  myGraph.AllpairsShortestPaths_MatrixMultiplication('a')
  myGraph.MinimumSpanningTree_Prim('a')
  myGraph.SingleSourceShortestPath_Dijkstra('a')
  # myGraph.Draw()

更多关于Python相关内容可查看本站专题:《Python正则表达式用法总结》、《Python数据结构与算法教程》、《Python Socket编程技巧总结》、《Python函数使用技巧总结》、《Python字符串操作技巧汇总》、《Python入门与进阶经典教程》及《Python文件与目录操作技巧汇总》

希望本文所述对大家Python程序设计有所帮助。

标签:
Python,图算法

金狮镖局 Design By www.egabc.com
金狮镖局 免责声明:本站文章均来自网站采集或用户投稿,网站不提供任何软件下载或自行开发的软件! 如有用户或公司发现本站内容信息存在侵权行为,请邮件告知! 858582#qq.com
金狮镖局 Design By www.egabc.com

评论“Python图算法实例分析”

暂无Python图算法实例分析的评论...

《魔兽世界》大逃杀!60人新游玩模式《强袭风暴》3月21日上线

暴雪近日发布了《魔兽世界》10.2.6 更新内容,新游玩模式《强袭风暴》即将于3月21 日在亚服上线,届时玩家将前往阿拉希高地展开一场 60 人大逃杀对战。

艾泽拉斯的冒险者已经征服了艾泽拉斯的大地及遥远的彼岸。他们在对抗世界上最致命的敌人时展现出过人的手腕,并且成功阻止终结宇宙等级的威胁。当他们在为即将于《魔兽世界》资料片《地心之战》中来袭的萨拉塔斯势力做战斗准备时,他们还需要在熟悉的阿拉希高地面对一个全新的敌人──那就是彼此。在《巨龙崛起》10.2.6 更新的《强袭风暴》中,玩家将会进入一个全新的海盗主题大逃杀式限时活动,其中包含极高的风险和史诗级的奖励。

《强袭风暴》不是普通的战场,作为一个独立于主游戏之外的活动,玩家可以用大逃杀的风格来体验《魔兽世界》,不分职业、不分装备(除了你在赛局中捡到的),光是技巧和战略的强弱之分就能决定出谁才是能坚持到最后的赢家。本次活动将会开放单人和双人模式,玩家在加入海盗主题的预赛大厅区域前,可以从强袭风暴角色画面新增好友。游玩游戏将可以累计名望轨迹,《巨龙崛起》和《魔兽世界:巫妖王之怒 经典版》的玩家都可以获得奖励。