复杂网络MATLAB工具箱

最新推荐文章于 2024-07-08 20:49:24 发布

JQKAgo

最新推荐文章于 2024-07-08 20:49:24 发布

阅读量2.8k

点赞数

分类专栏： MATLAB 复杂网络文章标签： matlab 复杂网络

复杂网络同时被 2 个专栏收录

2 篇文章 1 订阅

订阅专栏

MATLAB

1 篇文章 0 订阅

订阅专栏

幂律分布：

连续型：以f(x)表示某一数量指标的发生次数，若 f(x)=f(x）^-a ，就称为幂律分布。

离散型：若p(k)为离散型随机变量的概率分布，若p(k)~ck^-(a+1) ，则称p(k)为幂律分布。

p(k)=c/k^-a , logp(k)=logc-alogk

在双对数坐标系下，该分部呈现为一条斜率为负幂指数的直线。

基本规律总结起来：只有少数节点才会被经常使用，大部分的节点很少被使用。

一个网络的度分布服从幂律分布，直观上看就是少部分节点度极大，与许多节点相连，而大部分节点度都比较小，这种网络就叫做无标度网络。

复杂网络MATLAB工具包：

http://www.levmuchnik.net/常用的一些关于复杂网络的MATLAB程序。

http://www.levmuchnik.net/Content/Networks/NetworkPackageTutorial.html 复杂网络包的教程

http://www.levmuchnik.net/Content/Networks/ComplexNetworksPackage.html#Download 复杂网络MATLAB工具包的下载

教程1：

http://www.levmuchnik.net/Content/Networks/Tutorials/Tutorial_1/NetworkPackageTutorial_1.html

展示了如何生成一个随机无标度网络图，基本结构分析，包括度分布、聚类系数等。观察两种类型的攻击对网络的影响：随机攻击（所有的几点有相同的概率从网络中删除）、有目标地选择连接度最大的节点（节点度最大的节点从网络中删除）。

第二中情况下，网络几乎很快崩溃，第一种情况好一些。

教程1的相关MATLAB程序：

%% Tutorial 01

% The following scripts is a first phase of the Complex Networks Package tutorial.

% It demonstrates genration of random scale-free network and some basic analysis of it.

%% Genrate the network:生成一个网络

NumberOfNodes = 10000; % Number of nodes节点的个数

Alpha = -2.2; % Alpha of the scale-free graph 无标度网络的幂律指数

%define node degree distribution: 定义节点度分布

XAxis = unique(round(logspace(0,log10(NumberOfNodes),25)));

YAxis = unique(round(logspace(0,log10(NumberOfNodes),25))).^(Alpha+1);

% create the graph with the required node degree distribution: 根据所需节点度分布创建图

Graph = mexGraphCreateRandomGraph(NumberOfNodes,XAxis,YAxis,1);

%% Some basic statistics:一些基本特性

%Number of nodes in graph:图的节点的个数

disp(sprintf(‘Number of Nodes: %d’,GraphCountNumberOfNodes(Graph)));

%Number of links in graph:图的连边的个数

disp(sprintf(‘Number of Links: %d’,GraphCountNumberOfLinks(Graph)));

%Average degree:平均度

Degrees = GraphCountNodesDegree(Graph);

disp(sprintf(‘Average Node Degree: %2.2f’,mean(Degrees(:,2))));

%Find fraction of reciprocal links:找出相互连接的部分

Reciprocal = GraphCountUnderectionality(Graph);

disp(sprintf(‘Fraction of reciprocal links: %2.2f%%’,Reciprocal.DoubleConnectivityFraction*100));

% Clustering coefficient:聚类系数

CC = mexGraphClusteringCoefficient(Graph);

disp(sprintf(‘Average Clustering Coefficient: %3.3f%%’,CC.C));

%% Node Degree Distribution:度分布（度分布是复杂网络的最重要的一个需要分析的属性）

h1 = figure;

% incoming:入度

[y x] = hist(Degrees(:,2),unique(Degrees(:,2)));

loglog(x,y/sum(y),’*r’);

hold on

% outgoing：出度

[y x] = hist(Degrees(:,3),unique(Degrees(:,3)));

loglog(x,y/sum(y),’dg’);

% expected distribution:预期的分布

loglog(XAxis,YAxis/sum(YAxis),’:b’);

xlabel(‘k,Degree’);

ylabel(‘P(k)’);

title(‘Node Degree Distribution’);

legend({‘Incoming’,’Outgoing’,’Expected’});

%% Clustering Coefficient Distribution：聚类系数分布（高聚类特性是现实世界中无标度网络的另一个重要特性）

h2= figure;

% direct

CCin = mexGraphClusteringCoefficient(Graph,[],’direct’);

[y x] = hist(CCin.NodeClusteringCoefficient,linspace(0,1,25));

plot(x(x>0),y(x>0)/sum(y),’*r’);

hold on;

% inverse

CCout = mexGraphClusteringCoefficient(Graph,[],’inverse’);

[y x] = hist(CCout.NodeClusteringCoefficient,linspace(0,1,25));

plot(x(x>0),y(x>0)/sum(y),’dg’)

xlabel(‘CC, clustering coefficient’);

ylabel(‘P(CC)’);

title(‘Clustering coefficient distribution (CC>0)’);

legend({‘Direct’,’Inverse’});

%% Clustering Coefficient Dependence of Degree：聚类系数与度之间的关系（一个节点的聚类系数和节点度之间有什么关系？节点的平均聚类系数与度之间的关系分析）

h3 = figure;

% direct

loglog(CCin.k,CCin.Ck,’*r’);

hold on;

% inverse

loglog(CCout.k,CCout.Ck,’dg’);

xlabel(‘k, Degree’);

ylabel(‘<CC(k)>, clustering coefficient’);

title(‘Average clustering coefficient as a function of degree’);

legend({‘Direct’,’Inverse’});

%% Components组件（找出所有的强连接组件，）

% remove random node untill the graph breaks apart:随机删除节点直到图分裂

TempGraph=Graph;

NodesToRemovePerStep =1;

NumbersOfNodes = [];

NumbersOfLinks = [];

NumbersOfComponents = [];

LargestClusterSizes = [];

SecondLargestClusterSizes = [];

RemainingNodes = 1:NumberOfNodes;

while ~isempty(RemainingNodes)

NodeIndecesToRemove = unique(round(rand(NodesToRemovePerStep,1)*(numel(RemainingNodes)-1))+1);

NodesToRemove = RemainingNodes(NodeIndecesToRemove);

RemainingNodes = setdiff(RemainingNodes,NodesToRemove);

TempGraph = mexGraphNodeRemove(TempGraph,NodesToRemove);

NumbersOfNodes(end+1) = GraphCountNumberOfNodes(TempGraph);

NumbersOfLinks(end+1) = GraphCountNumberOfLinks(TempGraph);

if NumbersOfLinks(end)>0

Components = mexGraphConnectedComponents(TempGraph);

NumbersOfComponents(end+1) = numel(Components);

ComponentsSizes = sort(cellfun(‘length’,Components),’descend’);

if ~isempty(ComponentsSizes)

LargestClusterSizes(end+1) = ComponentsSizes(1);

else

LargestClusterSizes(end+1) = 0;

end

if numel(ComponentsSizes)>1

SecondLargestClusterSizes(end+1) = ComponentsSizes(2);

else

SecondLargestClusterSizes(end+1) = 0;

end

else

NumbersOfComponents(end+1) = 0;

LargestClusterSizes(end+1) = 0;

SecondLargestClusterSizes(end+1) = 0;

end

h4 = figure;

plot(NumbersOfComponents,’r’);

hold on;

h5 = figure;

plot(NumbersOfNodes,’r’);

hold on;

h6 = figure;

plot(NumbersOfLinks,’r’);

hold on;

h7 = figure;

plot(SecondLargestClusterSizes,’r’);

hold on;

h8=figure;

plot(LargestClusterSizes,’r’);

hold on;

%% remove most connected nodes (with heighest outgoing degree)：删除出度最大的节点

TempGraph=Graph;

NodesToRemovePerStep =1;

NumbersOfNodes = [];

NumbersOfLinks = [];

NumbersOfComponents = [];

LargestClusterSizes = [];

SecondLargestClusterSizes = [];

RemainingNodes = 1:NumberOfNodes;

while ~isempty(TempGraph.Data)

Degrees = GraphCountNodesDegree(TempGraph);

[OutDegrees SortOrder]=sort( Degrees(:,3),’descend’);

NodesToRemove = Degrees(SortOrder(1:min([numel(SortOrder) NodesToRemovePerStep])));

TempGraph = mexGraphNodeRemove(TempGraph,NodesToRemove);

NumbersOfNodes(end+1) = GraphCountNumberOfNodes(TempGraph);

NumbersOfLinks(end+1) = GraphCountNumberOfLinks(TempGraph);

if NumbersOfLinks(end)>0

Components = mexGraphConnectedComponents(TempGraph);

NumbersOfComponents(end+1) = numel(Components);

ComponentsSizes = sort(cellfun(‘length’,Components),’descend’);

if ~isempty(ComponentsSizes)

LargestClusterSizes(end+1) = ComponentsSizes(1);

else

LargestClusterSizes(end+1) = 0;

end

if numel(ComponentsSizes)>1

SecondLargestClusterSizes(end+1) = ComponentsSizes(2);

else

SecondLargestClusterSizes(end+1) = 0;

end

else

NumbersOfComponents(end+1) = 0;

LargestClusterSizes(end+1) = 0;

SecondLargestClusterSizes(end+1) = 0;

end

figure(h4)

plot(NumbersOfComponents,’g’);

xlabel(‘Step’);

ylabel(‘Number of components’);