AI-Toolbox

Why we included this project

AI-Toolbox is a C++ library for sequential decision problems where an agent has to act under uncertainty: Markov decision processes, partially observable MDPs, and bandit algorithms. Python bindings let you prototype against native Python models, including OpenAI gym environments, before moving to C++ for speed. The interface is deliberately small: you supply your own model by implementing a handful of methods, and the bundled solvers take it from there. It also parses the Cassandra POMDP format, so existing benchmark problems carry over. Newcomers get simple tutorials and a JMLR paper to lean on, though this is a research toolbox rather than a turnkey product: expect to write your own model code, but the breadth of algorithms and utilities covers a lot of ground.