Stratified resampling of phytosociological databases: some strategies for obtaining more representative data sets for classification studies