reallift 0.1.0

Causal Inference Library for Lift Measurement

🚀 RealLift

Causal Inference Library for Lift Measurement

RealLift is a powerful Python library designed to help data scientists and marketers measure the true impact of their interventions (treatments) using advanced causal inference techniques, such as GeoLift, Synthetic Control, and Placebo Testing.

---

✨ Features

• 🎯 Geo-Splitting: Automatically find the best treatment and control groups based on historical correlation.
• 📈 Synthetic Control: Build counterfactuals using a weighted combination of control regions.
• 🧪 Significance Testing: Robust bootstrap-based confidence intervals and p-values.
• 🛡️ Placebo Tests: Validate your model's reliability by running "fake" experiments on control groups.
• ⏳ Duration Estimation: Calculate the necessary experiment duration based on MDE (Minimum Detectable Effect).
• 📊 Professional Visualizations: Built-in plotting for experiment results and validation.

---

🚀 Installation

Install it directly from source:

pip install .

Coming soon to PyPI: pip install reallift

---

⚡ Quick Start

1. Run a Complete Geo Experiment

from reallift import run_geo_experiment

result = run_geo_experiment(
    filepath="your_data.csv",
    date_col="date",
    treatment_start_date="2025-05-01",
    n_treatment=1,
    mde=0.1
)

# Access results
print(f"Observed Lift: {result['results'][0]['synthetic']['lift_mean_abs']:.4f}")

2. Manual Geo Splitting

from reallift import find_best_geo_split

best_split = find_best_geo_split(
    filepath="historical_data.csv",
    date_col="date",
    n_treatment=2
)
print(best_split)

---

📖 Examples

For a deep dive into the library's capabilities, check out our demonstration notebook.

---

🛠️ Requirements

• Python 3.8+
• Pandas, Numpy, Scikit-Learn, Scipy, Matplotlib, CVXPY

---

📄 License

MIT License. See LICENSE for details.

---

Developed with ❤️ by Roberto Junior