Summary
Scikit-decide is an open-source library for modeling and solving planning, scheduling and reinforcement learning problems within a common API which helps break technical silos between different decision-making communities and enables seamless benchmarking of different approaches. For instance, one can solve PDDL problems with both classical planning (via a bridge to Unified Planning) and reinforcement learning (via a bridge to RLlib) solvers with very few lines of code, and compare the different solutions. Thinking of both algorithm providers and solver users, the library’s class hierarchy has been designed to ease the integration of new domains and algorithms depending on their distinctive features (e.g. partially vs fully observable states, deterministic vs probabilistic state transitions, single vs multi agents, simulation-based vs formal transition models, etc.).
With more than 125k total downloads and 200 downloads per day on PyPi, the library is gaining traction in the global sequential decision-making landscape, including practitioners and researchers. It is officially sponsored by ANITI (the Artifical and Natural Intelligence Toulouse Institute) and is the main host for the research algorithms produced in the Horizon Europe’s TUPLES project (Trustworthy Planning and Scheduling with Learning and Explanations).
The half-day tutorial will show how to model and solve the same problems using algorithms from different communities, and how to extend the libraries with new domains and solvers in a few lines of code. It will alternate presentations and live Python coding sessions.
Agenda
- Introduction (15 mn): General concepts of the library: domains, solvers, spaces, hub, features
- Part I (90mn): Solving domains (aka problems) with auto-selected compatible solvers
- Notebook I: Solving control problems with reinforcement learning, and genetic programming solvers
- Notebook II: Solving scheduling problems with constraint programming, operation research, and reinforcement learning solvers
- Notebook III: Solving PDDL problems with classical planning, and reinforcement learning solvers
- Part II (60mn): Implementing your own domains and solvers
- Notebook IV: Implementing a scikit-decide domain for RDDL problems
- Notebook V: Implementing a scikit-decide solver embedding the JaxPlan and GurobiPlan planners and solving RDDL-based scikit-decide domains
- Conclusion (15mn): Applications, contribution guidelines, and future developments