Playing in continuous spaces: some analysis and extension of population-based incremental learning