This module is a simple wrapper for the rlgym environment that makes it compatible with the distrib-rl project.
This library exists to handle two special concerns:
- Compatibility with the distrib-rl config system
- Compatibility with the latest gym environment specification (v0.25.x at time of writing)
For the moment this library should be considered highly unstable, as our first
official release is not yet planned. As the various components of distrib-rl
stabilize, so too with this library, however as long as this library is
maintained it will continue to be updated to track the latest versions of gym
(or whatever reinforcement learning environment specification distrib-rl
requires) and distrib-rl
Once published, this library will make strict use of the semver 2.0.0 versioning specification, and those depending on this library in their own projects should take care to set their version ranges appropriately.
Due to the goals stated above, this project is at the mercy of the gym API with respect to its stability. That is, we cannot meet the goal of tracking the latest gym environment version if we limit our ability to make breaking changes to this environment. As a result, expect that breaking changes may be made to this library at any time. However as stated above, we will always communicate the presence of these changes in the way prescribed by the semver versioning standard.
For the moment the only portions of our API that we consider to be public are
the (currently undocumented) definitions of the dict structures for distrib-rl
environment config, and the portions of this API that are extensions of the Gym
Env
class.
As we move closer to an official public release, we will endeavor to document the config API via JSONSchema or similar.