Reinforcement Learning Python Library for Simplified AI Development

Many medical practices can be modeled using Markov decision processes (MDPs); furthermore, by embedding MDPs in Artificial Intelligence (AI) tools, machine learning is increasingly being used for clinical decision-making. Most MDP implementations in AI use free-ware public libraries, such as MDP toolbox, built for the Python programming language, which at the time of writing is the dominant language used for AI applications. However, algorithms in existing libraries can be further optimized to enable both faster and more complex clinical decisions by leveraging latest developments in Reinforcement Learning (RL). RL achieves an objective, such as medical treatment, by maximizing a reward function determined by interacting with a stochastic environment in which specific sequences of actions may have variable rewards. In this paper we present our RL optimized MDP toolbox that not only allows for more advanced AI medical usage but also significantly reduces the learning curve necessary for RL utilization. Major beneficiaries of this research will be the doctors and medical researchers who employ advanced AI in their decision-making processes.