Acme
An open-source Python reinforcement learning library with reference agents, algorithm baselines, and support for single-stream and distributed research.
Acme is a Python reinforcement learning library for researchers who need reference agents and reusable components for their own experiments. Its agents provide baselines for comparing algorithm performance, while the underlying building blocks let researchers develop new approaches without starting from scratch.
The library supports both single-stream and distributed agents. That makes it relevant to research at different scales, with components designed to work across those settings. Its focus is on readable agent implementations that researchers can study and adapt, alongside efficiency when running experiments.
Acme includes working examples of different agents in a variety of environments, plus a quickstart notebook. These give prospective users concrete implementations to assess before choosing it as a foundation for their research. It's open source under the Apache 2.0 license. Install dm-acme with the JAX or TensorFlow extras required by the chosen agent. Example environments are separate extras; a Python virtual environment helps isolate their dependencies.
