Coach RL , this is an open-source framework for training and evaluation of reinforcement learning agent. specific contents:
- 1.0.0 version launched fusion of the "update" and "more powerful" algorithm, while increasing the availability of RL Coach of the API. RL Coach 1.0.0 add a total of 27 kinds of reinforcement learning model, specifically, the feedback loop is driven by these models to guide them to achieve the target of incentives, and support the use Coach as API Python library. In addition, improved documentation and unspecified bug fixes and general performance enhancements.
- RL Coach 1.0.0 played a good batch of intensive study, and enable the so-called closed policy evaluation (OPE), the robustness of the learning strategies (ie set of rules, which specify test of AI agents in all contingencies should do ) use of other policies on the basis of the data obtained.
- In addition, it adds support for several new reinforcement learning agents, including the Sample Efficient preference Experience with the Actor-Critic Replay (Acer), Soft the Actor-Critic (SAC) and double-deep certainty delay tactics gradient (TD3).
The official pointed out that it only been tested on Ubuntu 16.04 LTS and Python3.5. It is used with the same name OpenAI Gym OpenAI.
Details are explained