poke-env
Basic Information
poke-env is a Python interface aimed at developers and researchers who want to create and train reinforcement learning agents that battle on Pokemon Showdown. The repository provides programmatic access to run automated matches on the Pokemon Showdown platform so agents can be trained, evaluated, and benchmarked in competitive battles. It is intended as a development tool to bridge RL training workflows and the Pokemon Showdown environment, enabling automated self-play or matches against remote opponents. The project targets users building bots and experiments rather than end users, offering an environment for iterating on agent policies, testing strategies in a standardized battler, and collecting experience for training algorithms.
Links
Stars
385
Language
Github Repository
Categorization
App Details
Features
Provides a Python API focused on reinforcement learning for Pokemon battles. Enables automated execution of matches on the Pokemon Showdown platform so agents can be repeatedly trained and evaluated. Offers programmatic hooks for agents to receive game state information and submit actions during battles. Supports both self-play and matches versus external opponents to generate training data and performance metrics. Designed to integrate into training loops and experimental workflows so developers can run many episodes and compare agent policies under consistent conditions.
Use Cases
It helps researchers and developers accelerate development of competitive Pokemon agents by providing a ready-made bridge between RL algorithms and the Pokemon Showdown battler. By automating match execution and exposing game state and action interfaces, the repository lets teams collect large amounts of training experience, perform reproducible evaluations, and iterate on policy designs efficiently. It is useful for experimentation, benchmarking strategies, creating self-play pipelines, and serving as an educational platform to demonstrate reinforcement learning in a complex, adversarial game environment.