deep-trading-agent

Why we included this project

This project is a compact reference for anyone curious about applying deep reinforcement learning to trading. It models trading as a Markov decision process: the agent holds neutral, long, or short positions and learns to maximize accumulated reward, with a DeepSense-style network approximating the Q-function over per-minute Bitcoin price data. The repo covers the whole pipeline, from pulling and sampling exchange transaction history to preprocessing prices and training with TensorFlow while watching progress in TensorBoard, and it ships a Docker image so you can start a training run without installing the Python 2.7 dependencies yourself. Keep in mind this is a research implementation, not a production trading system, so treat it as a learning base and a place to experiment rather than something to hook into live order execution.