Playing with actor critic deep reinforcement learning models for automating and optimizing stock trading strategies to maximize profit in a custom OpenAI gym. We will use pretrained models using the stable_baselines library (A2C, PPO2, TRPO) and a custom DDPG model in Keras (buggy)
- Stock Trading with RL.ipynb: Jupyter Notebook for interacting with the different components
- env.py: StockTradingEnv OpenAI gym environment, where we define the observation space, agent actions (BUY, SELL, HOLD and percentage of shares (continuous action space)).
- graph.py: Used to render live trades from the agent
- agent.py: Implementation of a DDPG (Deep Deterministic Policy Gradient) RL agent.
- models.py: Contains the Actor and Critic models used by the DDPG agent (actor maps states to actions, critic returns Q value of the state action mapping)
- utils.py: Utility functions used by the project