Enhancements in EM implementation

[GustavoRodovalho]

Hello, @Ceyron!

I was watching your very informative youtube video about the EM algorithm for the multivariate case of a GMM (em_multivariate_gmm.py) and I'm interested in exploring two potential enhancements:

1. imputation of missing values by the algorithm (instead of using the mean and other regression methods), and
2. flexibilize the type of covariance matrices (add the diagonal and spherical cases)

For a quick context, I'm trying to use the EM algorithm to find the most representative samples of each cluster and my approach is based on another method called Self-Organizing Maps, more specifically the adapted version called Self-Organizing Mixture Models.

Hope this issue reaches you!

Best regards,

Gustavo Rodovalho.