Effective Reinforcement Learning for Reasoning in Language Models