From 2918c7fc9a4db37f03a0a94d455e393b26fc5e72 Mon Sep 17 00:00:00 2001
From: Mo Chen <sth4nth@gmail.com>
Date: Mon, 7 Mar 2016 17:12:13 +0800
Subject: [PATCH] removed mixMnEm

---
 chapter09/mixBernEm.m |  2 +-
 chapter09/mixMnEm.m   | 48 -------------------------------------------
 2 files changed, 1 insertion(+), 49 deletions(-)
 delete mode 100644 chapter09/mixMnEm.m

diff --git a/chapter09/mixBernEm.m b/chapter09/mixBernEm.m
index 6c94a40..65612e4 100644
--- a/chapter09/mixBernEm.m
+++ b/chapter09/mixBernEm.m
@@ -1,7 +1,7 @@
 function [label, model, llh] = mixBernEm(X, k)
 % Perform EM algorithm for fitting the Bernoulli mixture model.
 % Input: 
-%   X: d x n data matrix
+%   X: d x n binary (0/1) data matrix 
 %   k: number of cluster
 % Output:
 %   label: 1 x n cluster label
diff --git a/chapter09/mixMnEm.m b/chapter09/mixMnEm.m
deleted file mode 100644
index d22b944..0000000
--- a/chapter09/mixMnEm.m
+++ /dev/null
@@ -1,48 +0,0 @@
-function [label, model, llh] = mixMnEm(X, k)
-% Perform EM algorithm for fitting the multinomial mixture model.
-% Input: 
-%   X: d x n data matrix
-%   k: number of cluster
-% Output:
-%   label: 1 x n cluster label
-%   model: trained model structure
-%   llh: loglikelihood
-% Written by Mo Chen (sth4nth@gmail.com).
-%% initialization
-fprintf('EM for mixture model: running ... \n');
-n = size(X,2);
-label = ceil(k*rand(1,n));  % random initialization
-R = sparse(1:n,label,1,n,k,n);
-tol = 1e-8;
-maxiter = 500;
-llh = -inf(1,maxiter);
-for iter = 2:maxiter
-    model = maximization(X,R);
-    [R, llh(iter)] = expectation(X,model);
-    if abs(llh(iter)-llh(iter-1)) < tol*abs(llh(iter)); break; end;
-end
-[~,label(:)] = max(R,[],2);
-llh = llh(2:iter);
-
-function [R, llh] = expectation(X, model)
-mu = model.mu;
-w = model.weight;
-n = size(X,2);
-logRho = X'*log(mu);
-logRho = bsxfun(@plus,logRho,log(w));
-T = logsumexp(logRho,2);
-llh = sum(T)/n; % loglikelihood
-logR = bsxfun(@minus,logRho,T);
-R = exp(logR);
-
-function model = maximization(X, R)
-n = size(X,2);
-nk = sum(R,1);
-w = nk/n;
-mu = bsxfun(@times, X*R, 1./nk);
-d = size(X,1);
-lambda = 1e-4;
-prior = (1/d)*ones(d,1);
-mu = bsxfun(@plus,(1-lambda)*mu,lambda*prior);
-model.mu = mu;
-model.weight = w;
\ No newline at end of file