These algorithms are designed to emulate a Reinforcement Learning (RL) environment in Java, which is capable of receiving actions from Python and returning the subsequent state and reward via the ZeroMQ Request/Response protocol. In Java, these algorithms extend the abstract. class SingleInstrumentRLAlgorithm. In Python, the environment algorithms extend RlAlgorithm.
The system integrates Python and Java to execute a backtest with an RL algorithm. The Python AI gym, based on stable baselines 3, is open to integration with other frameworks. The Java backtest connects each step with a ZeroMQ request-response protocol.
The algorithm is trained from Python and initiates the backtest with an rlHost and rlPort property. If a SingleInstrumentRLAlgorithm is launched and the host and port are configured, an OrdinaryBacktestRLGym is initiated. This will include a ZeroMqServer that responds to the Python gym with the subsequent state and reward.
Please note that it is not possible to initiate a backtest or live trading solely with Java when using a Reinforcement Learning (RL) algorithm. The Java launchers that have been defined are intended solely for testing purposes, utilizing a dummyAgent