Generating Random Variables from a Finite Mixture Model

We can use the finite mixture model to generate more random numbers estimated from the model. For example, using the geyser data:

load(geyser);

hist(geyser,20);

Computational Statistics Toolbox has the csfinmix() function for estimating the finite mixture model. Thus:

>> [p,mu,var]=csfinmix(geyser',[50,80],[1 1],[0.5 0.5],100,1e-5)

p =
    0.3076    0.6924

mu =
   54.2027   80.3603

var =
   24.5223   56.3646

We can use this model to generate more samples, let's try 500 more of them:

n=500;

r = rand(1,n);

ind1 = find(r <= p(1));

ind2 = find(r > p(1));

x(ind2)=normrnd(mu(2),sqrt(var(2)),size(ind2));

x(ind1)=normrnd(mu(1),sqrt(var(1)),size(ind1));

figure;

hist(x,20);

Compare the below figure to the figure above.