Estatísticas para Training tabular generative adversarial networks in a federated learning framework for generating realistic, non-sensitive data for Modatta: a case study of recommendation systems in hyperbolic space