Skip to content

QUICKSTART.md

Latest commit

 

History

History
160 lines (122 loc) · 3.42 KB

QUICKSTART.md

File metadata and controls

160 lines (122 loc) · 3.42 KB

QuantLab Quick Start Guide

🚀 Get Started in 5 Minutes

1. Activate Environment

cd /Users/zheyuanzhao/workspace/quantlab
source .venv/bin/activate

2. Choose a Configuration

Option A: Liquid Universe (Recommended)

cd qlib_repo/examples
uv run qrun ../../configs/lightgbm_liquid_universe.yaml

Option B: Fixed Dates (2024 only)

cd qlib_repo/examples  
uv run qrun ../../configs/lightgbm_fixed_dates.yaml

Option C: Full Universe (All stocks)

cd qlib_repo/examples
uv run qrun ../../configs/lightgbm_external_data.yaml

3. View Results

The backtest will output:

  • IC and Rank IC metrics
  • Portfolio analysis
  • Results saved to results/mlruns/

4. Visualize

Find your experiment ID from the output, then:

# Edit scripts/analysis/visualize_results.py
# Update EXP_DIR to your experiment ID

cd /Users/zheyuanzhao/workspace/quantlab
uv run python scripts/analysis/visualize_results.py

Output: results/visualizations/backtest_visualization.png

📊 What You'll See

Console Output

IC: 0.066
ICIR: 0.622
Rank IC: -0.006
Sharpe Ratio: 3.94
Max Drawdown: -39.19%

Visualization

  • IC over time (predictive power)
  • Rank IC over time (ranking ability)
  • Cumulative returns chart
  • Drawdown analysis
  • Daily returns distribution

🔍 Understanding the Results

Good Signs ✅

  • IC > 0.05: Model has predictive power
  • ICIR > 0.5: Predictions are consistent
  • Sharpe > 2: Good risk-adjusted returns

Warning Signs ⚠️

  • Rank IC ≈ 0: Model can't rank stocks well
  • Extreme returns: May indicate data quality issues
  • High drawdown: Needs better risk controls

📁 Where Are My Results?

results/
├── mlruns/489214785307856385/
│   ├── [run_id_1]/artifacts/
│   │   ├── pred.pkl              # Model predictions
│   │   ├── sig_analysis/         # IC analysis
│   │   └── portfolio_analysis/   # Backtest results
│   └── [run_id_2]/artifacts/
│       └── ...
└── visualizations/
    └── backtest_visualization.png

🔄 Next Steps

  1. Read the analysis: docs/BACKTEST_SUMMARY.md
  2. Try different models: Edit configs to use XGBoost, MLP
  3. Improve features: Add custom Alpha features
  4. Better universe: Filter by market cap, volume

⚙️ Common Commands

List all experiment runs

ls results/mlruns/489214785307856385/

Check MLflow UI (if installed)

cd results
mlflow ui
# Open http://localhost:5000

Clean old runs

rm -rf results/mlruns/.trash

Update data

python scripts/data/refresh_today_data.py

🐛 Troubleshooting

"Module not found: qlib"

uv pip install qlib

"Data not found"

Check: /Volumes/sandisk/quantmini-data/data/qlib/stocks_daily/

"Import error"

Make sure you're in project root when running scripts

Visualization fails

uv pip install matplotlib seaborn pandas

📚 Learn More

  • README.md - Full project overview
  • docs/BACKTEST_SUMMARY.md - Detailed analysis
  • docs/USE_QLIB_ALPHA158.md - Feature guide
  • PROJECT_STRUCTURE.md - Directory layout

💡 Tips

  1. Start with lightgbm_liquid_universe.yaml - it's the most realistic
  2. Check IC before trusting backtest returns
  3. Compare multiple runs to validate results
  4. Read the docs - lots of insights there!

Happy backtesting! 🎯