Created
January 23, 2013 08:44
-
-
Save shunsukeaihara/4603227 to your computer and use it in GitHub Desktop.
Mixture model for graph
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# -*- coding: utf-8 -*- | |
import math | |
import numpy as NP | |
from scipy.maxentropy import logsumexp | |
from operator import itemgetter | |
class NewmanEM: | |
def __init__(sel): | |
pass | |
def ll(self): | |
"""対数尤度の計算""" | |
res=0.0 | |
for i in range(self.size): | |
for r in range(self.groups): | |
sum=0.0 | |
for j in self.outEdge[i]: | |
sum+=self.theta[r,j] | |
if math.exp(self.q[i,r])>0.0: | |
res+=math.exp(self.q[i,r])*(self.pi[r]+sum) | |
return res | |
def initialize(self,gp,mat,th=0.0): | |
"""パラメータの初期化""" | |
self.groups=gp | |
self.size=mat.shape[0] | |
x=NP.random.random(gp) | |
self.pi=x/NP.sum(x) | |
x=NP.matrix(NP.random.random((gp,self.size))) | |
self.theta=x/x.sum(1) | |
self.pi = NP.log(self.pi) | |
self.theta = NP.log(self.theta) | |
self.q=NP.matrix(NP.zeros((self.size,gp))) | |
self.outEdge=[]#in edgeの保存 | |
for i in range(self.size): | |
row=mat[i] | |
self.outEdge.append([i for i in range(len(row)) if row[i]>th]) | |
self.inEdge=[]#out edgeの保存 | |
mat=mat.T | |
for i in range(self.size): | |
row=mat[i] | |
self.inEdge.append([i for i in range(len(row)) if row[i]>th]) | |
pass | |
def clustering(self,mat,vocab,gp): | |
pi,q,theta,ll=self.em(mat,gp,1000) | |
self.showGmResult(q,theta,vocab,gp) | |
def showGmResult(self,q,theta,vocab,gp): | |
for r in range(gp): | |
dic={} | |
for i in range(len(vocab)): | |
th=theta[r,i] | |
dic[vocab[i]]=math.exp(th) | |
ar=sorted(dic.items(), key=itemgetter(1), reverse=True) | |
print "group " + str(r) | |
for x in ar[0:10]: | |
print x[0] + ":" + str(x[1]) | |
print "----" | |
for r in range(gp): | |
for i in range(len(vocab)): | |
th=math.exp(q[i,r]) | |
dic[vocab[i]]=th | |
ar=sorted(dic.items(), key=itemgetter(1), reverse=True) | |
print "group " + str(r) | |
for x in ar[0:10]: | |
print x[0] + ":" + str(x[1]) | |
print "----" | |
def em(self,mat,gp,step): | |
"""EMアルゴリズムで計算""" | |
self.initialize_em(gp,mat) | |
for st in range(step): | |
qold=self.q.copy() | |
#E-step qを計算 | |
for i in range(self.size): | |
edge=self.outEdge[i] | |
for r in range(self.groups): | |
self.q[i,r]=self.pi[r] | |
for t in edge: | |
self.q[i,r]+=(self.theta[r,t]) | |
pass#edge | |
pass#j | |
self.q[i]=self.q[i]-logsumexp(self.q[i]) | |
pass | |
#M-step theta,piを計算 | |
for r in range(self.groups): | |
ar=[] | |
for i in range(self.size): | |
ar.append(self.q[i,r]+NP.log(len(self.outEdge[i]))) | |
for j in range(self.size): | |
edge=self.inEdge[j] | |
ar2=[] | |
for t in edge: | |
ar2.append(self.q[t,r]) | |
self.theta[r,j]=logsumexp(ar2)-logsumexp(ar) | |
pass#j | |
pass#i | |
tq=self.q.T | |
for i in range(self.groups): | |
self.pi[i]=logsumexp(tq[i])-math.log(self.size) | |
deltasq=0.0 | |
for i in range(self.size): | |
for j in range(self.groups): | |
deltasq+=((NP.exp(qold[i,j])-NP.exp(self.q[i,j]))**2) | |
delta=math.sqrt(deltasq) | |
print "iter " + str(st) + ",d " + str(delta) | |
if delta==0.0 or delta==float("inf"): | |
break | |
pass#end of iteration | |
return self.pi,self.q,self.theta,self.ll() | |
if __name__ == '__main__': | |
#隣接行列 | |
mat=NP.array([[0.0,1,1,1,1,0,0,0,0], | |
[1.0,0,0,0,0,1,1,1,0], | |
[1,0,0,1,1,0,0,0,0], | |
[1,0,1,0,1,0,1,0,0], | |
[1,0,1,1,0,0,0,0,1], | |
[0,1,0,0,0,0,1,1,1], | |
[0,1,0,0,0,1,0,1,0], | |
[0,1,1,0,0,1,1,0,0], | |
[0,0,0,0,1,1,0,0,0] | |
]) | |
em=NewmanEM() | |
pi,q,theta,ll=em.em(mat,2,1000) | |
print NP.exp(pi) | |
print NP.exp(q) | |
print NP.exp(theta) | |
print ll |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment