使用IGRAPH在Julia中生成随机配置模型图
最近,我开始使用朱莉娅(Julia)中的igraph来生成随机配置模型,因为LightGraphs在实现这些对象的时间内有问题(链接到与此对象有关的先前问题: rando_configuration_model(n,e)需要很长的时间/a>)。要生成此类图,我生成了一个vector e
(基于1个索引),从中,我生成了一个igraph对象g2
,如下所示,
using PyCall, Distributions
ig = pyimport("igraph")
α=0.625;N=1000;c=0.01*N;p=α/(α+c)
E = zeros(Int64,N)
test=false
while test == false
s=0
for i in 1:N
E[i] = rand(NegativeBinomial(α,p))
s += E[i]
end
if iseven(s) == true
test = true
else
end
end
g = ig.Graph.Realize_Degree_Sequence(E)
我的第一个问题与以下事实有关。 Python是基于0的索引。通过比较e
的组件与g
的度,从1个基本对象e
生成基于0的对象g
。这是正确的吗?
其次,我想强制执行随机配置图g
是简单的,没有自我循环或多重编号。 igraphs文档( https://igraph.org/c/doc/igraph-generators.html#igraph_realize_degree_sequence )表示,flag washe_edge_edge_dege_types:igraph_simple_sw
我无法完成工作找到在朱莉娅中使用它的语法。是否有可能在朱莉娅(Julia)中使用此标志?
Recently I started to use iGraph in Julia to generate random configuration models, since LightGraphs has a problem with time realization of these objects (link to a previous question related to this: random_configuration_model(N,E) takes to long on LightGraphs.jl). To generate such graphs, I generate a vector E
(1-based indexed) and from it I generate an iGraph object g2
as follows
using PyCall, Distributions
ig = pyimport("igraph")
α=0.625;N=1000;c=0.01*N;p=α/(α+c)
E = zeros(Int64,N)
test=false
while test == false
s=0
for i in 1:N
E[i] = rand(NegativeBinomial(α,p))
s += E[i]
end
if iseven(s) == true
test = true
else
end
end
g = ig.Graph.Realize_Degree_Sequence(E)
My first question is related to the fact that python is 0-based indexed. By comparison of the components of E
to the degrees of g
, it seems that ig.Graph.Realize_Degree_Sequence(E)
automatically convert the index bases, generating a 0 based object g
from a 1-based object E
. Is this correct?
Secondly, I would like to enforce the random configuration graph g
to be simple, with no self loops nor multi-edges. iGraphs documentation (https://igraph.org/c/doc/igraph-Generators.html#igraph_realize_degree_sequence) says that the flag allowed_edge_types:IGRAPH_SIMPLE_SW
does the job, but I am not able to find the syntax to use it in Julia. Is it possible at all to use this flag in Julia?
如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。

绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱,如果其他用户或者作者回复了您的评论,将不能在第一时间通知您!
发布评论
评论(2)
请小心LightGraph的
ROTAMY_CONFIGRUATON_MODEL
。上次我看时,它被损坏了,它没有均匀地进行样品,,然而,作者彻底拒绝修复它。从那时起,我不知道是否有任何变化。c/igraph的 >具有正确实现的方法,该方法均匀地进行了样品,称为
igraph_degseq_simple_no_no_multiple_uniform
,但由于某种原因,它尚未在Python中暴露出来...我们会尽快将其暴露。然后,您有两个选项:
“ Simple”
方法,并继续生成图形,直到获得简单的图形(用graph.is_simple()
)。这使用了存根匹配方法,并将完全均匀地采样。对于很大的程度,由于许多拒绝,将需要很长时间。请注意,此拒绝方法恰好是igraph_degseq_simple_no_no_multiple_uniform
insterments insterments(尽管位更快)。graph.realize_degree_seperence()
用给定的度序列创建一个图,然后使用graph.rewire()
重写,并使用足够多的重新布线步骤(在边缘数量至少几倍)。该方法使用保留学位的边缘开关,可以显示在大量开关的极限下均匀地采样。“ no_multiple”
python-rigraph中的方法将再次 均匀示例。看看“ nofollow noreferrer”>第2.1节的本文解释哪些技术可用于均匀抽样。
Be careful with LightGraph's
random_configruaton_model
. Last time I looked, it was broken, and it did not sample uniformly, yet the authors outright refused to fix it. I don't know if anything changed since then.C/igraph's
degree_sequence_game()
has a correctly implemented method that samples uniformly, calledIGRAPH_DEGSEQ_SIMPLE_NO_MULTIPLE_UNIFORM
, but for some reason it is not yet exposed in Python ... we'll look into exposing it soon.Then you have two options:
"simple"
method, and keep generating graphs until you get a simple one (test it withGraph.is_simple()
). This uses the stub-matching method, and will sample exactly uniformly. For large degrees, it will take a long time due to many rejections. Note that this rejection method exactly what theIGRAPH_DEGSEQ_SIMPLE_NO_MULTIPLE_UNIFORM
implements (albeit bit faster).Graph.Realize_Degree_Sequence()
to create one graph with the given degree sequence, then rewrite it usingGraph.rewire()
with a sufficiently large number of rewiring steps (at least several times the edge count). This method uses degree-preserving edge switches and can be shown to sample uniformly in the limit of a large number of switches.The
"no_multiple"
method in python-igraph will again not sample uniformly.Take a look at section 2.1 of this paper for a gentle explanation of what techniques are available for uniform sampling.
您正在阅读c igraph的文档。您需要阅读python文档 https://igraph.org/python/api/latest/igraph._igraph.graph.graphbase.html#degree_sequence 。因此:
我使用
“ no_multiple”
算法,用作“ vl”
算法假定连接的图形,图形中的某些节点可以为0。You are reading C docs of igraph. You need to read Python documentation https://igraph.org/python/api/latest/igraph._igraph.GraphBase.html#Degree_Sequence. So:
I used
"no_multiple"
algorithm, as"vl"
algorithm assumes connected graph and some of degrees of nodes in your graph can be 0.