TradingGym is a platform for automated optimal trading. It implements OpenAI Gym environment to train and test reinforcement learning agents. The environment is created from level II stock exchange data and takes into account commissions, bid-ask spreads and slippage (but still assumes no market impact).
git clone https://github.com/ksemianov/TradingGym
cd TradingGym
python3 -m pip install -e .
The platform supports level II market data in Plaza II format from MOEX. It expects hdf5 file where every key has data for a separate trading session. The example value for a key should look similar to this:
See RL notebook for examples of training and testing agents based on keras-rl.