The parallel implementation is not releasing memory... it keeps growing. After some googling, I found that the pool in ParallelEvaluator should be closed by calling pool.close() followed by pool.join()