Jumanji v0.2.1

Release Notes

This major release introduces a whole set of new environments, namely:

• Game2048: the classic 4x4 2048 game
• Minesweeper: the classic game from the video game
• RubiksCube: the standard puzzle
• JobShop: job-shop scheduling problem as another canonical CO problem
• Cleaner: a maze with multiple agents that spawn across the grid in order to clean all the tiles ASAP
• Connector: replacement of the previous Routing environment reimplemented for better efficiency and modularity.
• Maze: classic maze on a grid

This release also stabilizes the API with a cleaner definition of the environment specs, rendering implemented for all the environments (using matplotlib), and an actor-critic agent implement for training alongside environment-specific actor-critic networks for each environment.

What's Changed

• fix: pygame requirements by @wang-r-j in #72
• feat: internal update for v0.2.0 by @clement-bonnet in #73
• refactor: remove connect4 environment
• feat: implement training and environment-specific networks in jumanji.training by @clement-bonnet
• feat: replace routing environment with connector by @sash-a
• feat: implement rubiks_cube environment by @TristanKalloniatis
• feat: implement game_2048 environment by @OmaymaMahjoub
• feat: implement minesweeper environment by @TristanKalloniatis
• feat: implement job_shop environment by @dluo96
• refactor: knapsack, cvrp and tsp by @surana01
• feat: implement cleaner environment by @coyettev
• feat: implement maze environment by @PDuckworth
• feat: clean and augment snake with action mask by @clement-bonnet
• refactor: implement arbitrarily nested specs by @dluo96
• ci: drop support for python 3.7

New Contributors

• @wang-r-j made their first contribution in #72

Full Changelog: v0.1.6...v0.2.1