tensortrade

Why we included this project

TensorTrade lets you put a reinforcement learning agent in front of real trading decisions and find out whether it can beat a buy-and-hold baseline. Instead of wiring together gym-style code yourself, you assemble the pieces of a trading loop from composable components: environments, action schemes, reward functions, and data feeds. It comes with a default buy/sell/hold scheme, a position-based reward scheme, and a portfolio layer that tracks wallets and positions, so the simulation behaves more like live trading than a bare RL benchmark. The maintainers also publish their own experiments, training PPO agents on BTC/USD and reporting P&L against buy-and-hold, including how commissions eat into gains; that transparency is useful before you sink weeks into your own runs. It suits engineers who already know RL and want a documented, composable environment for quantitative trading research.