Dragonfly: a modular deep reinforcement learning library
This provides a tool for researchers and developers in reinforcement learning to reduce code maintenance and accelerate experimentation, though it is incremental as it builds on existing methods.
The authors tackled the challenge of facilitating experimentation in deep reinforcement learning by developing Dragonfly, a modular library that uses JSON serialization to enable easy swapping of components and parameter sweeps, and it performs competitively on standard benchmarks compared to existing literature.
Dragonfly is a deep reinforcement learning library focused on modularity, in order to ease experimentation and developments. It relies on a json serialization that allows to swap building blocks and perform parameter sweep, while minimizing code maintenance. Some of its features are specifically designed for CPU-intensive environments, such as numerical simulations. Its performance on standard agents using common benchmarks compares favorably with the literature.