The limitations of softmax in reinforcement learning - View it on GitHub