Partitioning streaming parallelism for multi-cores: a machine learning based approach