Poke-env: A Python Interface for Training Reinforcement Learning Pokémon Bots

Poke-env provides an environment for engaging in Pokémon Showdown battles with a focus on reinforcement learning.

It boasts a straightforward API for handling Pokémon, Battles, Moves, and other battle-centric objects, alongside an OpenAI Gym interface for training agents.

Attention

While poke-env aims to support all Pokémon generations, it was primarily developed with the latest generations in mind. If you discover any missing or incorrect functionalities for earlier generations, please open an issue to help improve the library.

User guide

• Getting Started
  • Installing poke-env
  • Configuring a Pokémon Showdown Server
  • Creating Agents
  • Configuring Showdown Players
  • Connecting Bots to Showdown
• Examples
  • Cross Evaluating Random Players
  • Creating a simple max damage player
  • Reinforcement learning with the OpenAI Gym wrapper
  • Adapting the max player to gen 8 OU and managing team preview
  • Creating a custom teambuilder
  • Connecting to showdown and challenging humans

Main modules documentation

• Battle objects
• The player object and related subclasses
  • Env player
  • Player
  • OpenAIGymEnv
  • Random Player
• The pokémon object
• The move object
• Other objects
  • Effect
  • Field
  • Move category
  • Pokemon gender
  • Pokemon Type
  • Side condition
  • Status
  • Weather
  • Z Crystal

On top of the main modules dedicated to building Pokémon Showdown bots, Poke-env encompasses standalone submodules to cater to various facets of Pokémon Showdown interactions:

Standalone submodules documentation

• Data - Access and manipulate pokémon data
• PS Client - Interact with Pokémon Showdown servers
• Teambuilder - Parse and generate showdown teams

Acknowledgements

Originated from a group project at École Polytechnique. Inspired partially by showdown-battle-bot and built upon Pokemon Showdown.

Data

Data adapted from Pokemon Showdown and team data sourced from Smogon forums’ RMT section.

License

Released under the MIT License.