Interpretable statistics for complex modelling: quantile and topological learning