What it does
FinRL is an open-source framework for financial reinforcement learning. It combines market environments, learning agents and financial applications in a data, training and backtesting workflow. This repository serves education and research prototyping; the foundation directs deployment work to the separate FinRL-X project.
FinRL's stock-trading tutorial is organized into three scripts: prepare historical data, train learning agents, then compare their backtests. The data stage adds indicators and separates training from later evaluation data. Trained models are saved for reuse, and the final stage produces a chart and comparisons with the Dow Jones index and a mean-variance allocation.
The training example makes several assumptions visible in code. You can change starting capital, initial shares, maximum trade size, buying and selling costs, and the chosen indicators. Flags select which of the five agent types to train; the example writes training logs as well as model files. Inspect those settings before interpreting a result.
The MIT-licensed framework is free. Clone it into a local Python environment and follow the repository's installation instructions; FinRL itself has no account requirement. External data services or a broker connection can have separate accounts and costs. Expect programming and machine-learning work: this is a research implementation, with FinRL-X maintained as a separate deployment-oriented project.
Strengths
- Separates data preparation, model training and historical evaluation
- Exposes capital, trading-cost and indicator settings in the example
- Saves trained agents and comparison results
Watch out for
- Adapting experiments requires Python and machine-learning knowledge
- This repository is scoped to education and research prototyping
Best for
- Researchers studying reinforcement learning for portfolio decisions
- Developers reproducing financial machine-learning experiments
Sources
- FinRL classic workflow and research scope (Oct 27, 2026)
- FinRL stock-trading training example and environment controls (Oct 27, 2026)
- FinRL MIT license (Oct 27, 2026)