Missing data imputation through GTM as a mixture of t-distributions