How unasthana/portfolioallocationdeeprl works
Project Type and Purpose
• From the README and file structure, this is a machine-learning/data-science project focused on portfolio allocation using a latency-optimized deep reinforcement learning (A2C) algorithm in a simulated OpenAI Gym environment.
• It spans a data pipeline (scraping → cleaning → preprocessing), model training/evaluation (Main.ipynb), and successive code optimizations (pure Python → Cython → Numba → multiprocessing/multithreading).
File Structure
Top-level directories:
• Data/ – input asset list (stock_tickers.xlsx)
• RawCode/ – initial versions of three scripts plus install_packages.py
• OptimizedCode/ – optimized versions of the same scripts, plus Main.ipynb for orchestration
Configuration/build files:
• install_packages.py (dependency installer)
• README.md (project description & methodology)
README for Additional Insights
• Detailed methodology: custom Gym environment, state/action spaces, reward logic.
• A2C architecture: actor & critic networks, policy/value updates.
• Optimization stages: Python-level refactoring, Cython compilation, Numba JIT, parallel execution.