hajirazin/learnfinance-models-ppo

LearnFinance PPO Model - v2026-02-27-e5345ed2cf47

Proximal Policy Optimization (PPO) portfolio allocation agent using dual forecasts (LSTM + PatchTST) as features.

Model Details

• Version: v2026-02-27-e5345ed2cf47
• Model Type: PPO (Proximal Policy Optimization) with dual forecasts
• Training Window: 2016-01-01 to 2026-02-27
• Symbols: 15 stocks

Components

• weights.pt - PPO actor-critic network weights
• scaler.pkl - PortfolioScaler for state normalization
• symbol_order.json - Ordered list of portfolio symbols

Metrics

• Policy Loss: -0.008436506443942449
• Value Loss: 47129.908203125
• Avg Episode Return: 0.17079946008368832
• Avg Episode Sharpe: 0.1145480840508518
• Eval Sharpe: 1.2736954447449413
• Eval CAGR: 0.3237192191683296
• Eval Max Drawdown: 0.14662845878963948

Usage

from brain_api.storage.ppo import PPOHuggingFaceModelStorage

storage = PPOHuggingFaceModelStorage(repo_id="hajirazin/learnfinance-models-ppo")
artifacts = storage.download_model(version="v2026-02-27-e5345ed2cf47")