Deep-RL-Stock-Portfolio-Management

Description

This is an Reinforcement Learning agent using advantage REINFORCE (Baird 1994) to manage a portfolio in the stock market.

For a detailed description, please see this Devpost link

The trained agent is able to learn a good policy and make profits compared to the general stock index.

Structure

agent.py contains the advantage REINFORCE RL agent. This agent also uses Experience Replay to stabilize training.

main.py contain training, validation, and testing

preprocess.py contains the pre-processing of financial data

stock_env.py contains the simulation environment of the stock market. This environment takes into account interest rate, inflation, ... It contains functions allowing the agent to sample episodes of historical stock data using the agent's policy.

visual_helpers.py contains helper functions to visualize the portfolio the agent is managing

rnn.py contains a pre-trained Gated Recurrent Unit that can predict stock prices. rnn_model contains a saved model of this GRU, and data contains the pre-processed data needed to train this model. Ideally, this prediction made by this RNN is used as input to the RL agent, although this feature is not implemented yet.

Running the model

run the bash script create_venv.sh to create the virtual requirement for this model then run python main.py to start training and testing the model