Deep Reinforcement Learning Trading Agent with Synthetic Data

This project implements a Bitcoin trading agent using Deep Q-Learning (DQN) enhanced with synthetic data containing rare events such as crashes, spikes, and volatility segments.

---

Contents

• data.py: Download and preprocess real and synthetic Bitcoin data.
• environment.py: Trading environment definition.
• agent.py: DQN agent implementation using PyTorch.
• train.py: Training the agent with and without synthetic data.
• evaluate.py: Agent evaluation.
• visualize.py: Visualization of training results.
• main.py: Main script running the complete training and evaluation pipeline.

---

Usage Instructions

1. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # Linux/macOS
   .venv\Scripts\activate     # Windows

2. Install the required packages:

   pip install -r requirements.txt

3. Run the main script:

   python main.py

Notes

-Uses the yfinance library to fetch historical Bitcoin data.
-Synthetic data is generated simulating prolonged crashes, flash crashes, and volatile periods.
-The model is trained with Deep Q-Learning via PyTorch.
-Provides visualization of agent performance during training.
-Suitable for experimentation with reinforcement learning on financial data.