Relabelling algorithms for mixture models with applications for large data sets