Deep learning for molecular generation and optimization - a review of the state of the art