1. Matlab provides a class 'gmdistribution' for GMM高斯混合模型.
2. If you want to implement by yourself.There some tricks for implementing
this kind of algorithms numerically stable. For example:try to maximize log-
likelihood rather than the likelihood so that 连乘 becomes summation. When
computing the log-likelihood, you may need to subtract the maximal values
and put it back to prevent underflow.
