Next Financial Decision Model

This project utilizes reinforcement learning to develop a trading agent capable of operating in a financial environment. The implementation leverages the tf-agents library from TensorFlow for efficient and effective reinforcement learning algorithms.

Project Structure

• agents/: Contains the implementation of the PPO agent.
• checkpoints/: Contains the saved model checkpoints.
• data/: Contains the financial data used for training and evaluation.
• environments/: Contains the implementation of the financial environment.
• experiments/: Contains the scripts to run experiments.
• models/: Contains the model definitions.

Branches

The main branch is configured to run on Apple's M-series chips. The cuda12-version branch is configured to run on devices with CUDA 12.3. Both branches are up to date with the latest developments.

Setup

pip install pipenv
pipenv install

Train

Run and continue from last checkpoint

pipenv run python3 main.py

Run and clean all previous checkpoints

pipenv run python3 main.py --clean_run