Reinforcement Learning Stock Trading Agent

This project contains a reinforcement learning (RL) agent designed to trade stocks in a custom stock trading environment. The agent is trained using historical stock data with the goal of maximizing profit. The code provided in main.py demonstrates how to train the agent, execute trades, and visualize the results.

Additionally, this project includes a script for scraping stock data from Yahoo Finance (yfinance_scraper.py) and an example of an ensemble approach for trading using multiple RL algorithms (Ensemble.py).

Table of Contents

1. Requirements
2. Installation
3. Usage
4. Scraping Stock Data
5. Ensemble Trading
6. Customizing the Training
7. License

Requirements

• Python 3.x
• pandas
• torch
• ta (Technical Analysis Library)
• numpy
• gym-anytrading
• stable-baselines3
• enum
• yfinance

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/rl-stock-trading-agent.git
   cd rl-stock-trading-agent
   

2. Install the required dependencies:

   pip install -r requirements.txt
   

Usage

1. Prepare your stock data in CSV format with columns: Date, Open, High, Low, Close, and Volume.

2. Train the agent and visualize the results using the provided main.py script:

   python main.py --stock path/to/your/stock_data.csv
   

3. The training results and performance visualization will be displayed upon completion.

Scraping Stock Data

To scrape stock data from Yahoo Finance, use the yfinance_scraper.py script. The script saves stock data for Information Technology companies in the S&P 500 index in the data/ directory.

To run the script:

python yfinance_scraper.py

Ensemble Trading

The Ensemble.py script demonstrates an ensemble approach for stock trading using multiple reinforcement learning algorithms: PPO, A2C, and DQN. The script trains the models and performs trading based on a majority vote of their actions.

To run the ensemble script:

python Ensemble.py --stock path/to/your/stock_data.csv

Customizing the Training

You can modify the training parameters by passing additional arguments to main.py or Ensemble.py. These arguments should be defined in the parse_args function located in the parse_args.py module.

Example:

python main.py --stock path/to/your/stock_data.csv --window-size 60 --epochs 150 --hidden-nodes 240 --max-shares 200 --learning-rate 5e-4 --use-lr-scheduler False