reallift 0.1.3

Causal Inference Library for Lift Measurement

RealLift

Causal Inference Library for Lift Measurement & Design of Experiments

RealLift is an advanced Python library engineered to assist data scientists and analysts in reliably measuring the true incremental impact of interventions through rigorous causal inference methodologies, such as GeoLift, Synthetic Control Estimation, and Placebo Testing.

---

Capabilities

• Design of Experiments (Geo-Splitting): Algorithmically identifies structural clusters and mathematically selects the optimal treatment and control regions based on ElasticNet feature selection and convex proximity matrices.
• Synthetic Control Measurement: Formulates robust counterfactual interventions by mapping temporal correlations across a predefined array of donor regions via constrained Convex Optimization (cvxpy).
• Time Series Cross-Validation: Ensures predictive validity of counterfactuals via Historical Simulation, isolating definitive Out-Of-Fold (OOF) $R^2$ and MAPE limits prior to experiments.
• Duration & Statistical Power: Estimates predictive power dynamically over time streams to establish strict Minimum Detectable Effect (MDE) bounds before test implementation.
• Significance & Placebo Testing: Empirically defends the analytical conclusions through non-parametric bootstrap sampling and randomized spatial placebo permutations to comprehensively evaluate the null hypothesis.

---

Installation

RealLift is securely distributed through PyPI for production environments:

pip install reallift

Alternatively, obtain the latest development snapshot directly from the source repository:

pip install git+https://github.com/RobertoJuniorWXYZ/RealLift.git

---

Quick Start Guide

1. Requirements & Design (Pre-Test Phase)

Before executing a field intervention, analyze the underlying baseline correlation to discover optimal clusters and project the duration strictly necessary to capture a target Minimum Detectable Effect (MDE).

from reallift import run_geo_requirements

# Identify structural blocks and validate exposure durations
summary = run_geo_requirements(
    filepath="historical_data.csv",
    date_col="date",
    n_treatment=1,
    mde=0.015,
    max_days=[21, 60],
    n_folds=5,
    verbose=True
)

2. Intervention Measurement (Post-Test Phase)

Following the completion of an intervention, apply the algorithmic pipeline encompassing validation constraint-checking, Synthetic Control extraction, and empirical Placebo diagnostics.

from reallift import run_geo_experiment

# Execute the complete analytical pipeline
result = run_geo_experiment(
    filepath="experiment_data.csv",
    date_col="date",
    treatment_start_date="2025-05-01",
    n_treatment=1,
    mde=0.015,
    max_days=[21, 60],
    n_folds=5,
    random_state=42,
    verbose=True
)

# Extract total absolute impact estimates
print(f"Incremental Lift (abs): {result['results'][0]['synthetic']['lift_total']:.2f}")

---

Examples & Application

For a comprehensive methodological demonstration concerning correlation assumptions, feature engineering operations, and diagnostic evaluation limits, refer to the Jupyter notebooks mapped under the examples/geotests/ directory within the primary repository.

---

Systems & Dependencies

• Platform Target: Python 3.8+
• Mathematics: cvxpy (Core algorithmic solver for constraints)
• Data Engineering: pandas, numpy, scikit-learn, scipy
• Plotting Engines: matplotlib, seaborn

---

License

MIT License. Navigate to the LICENSE file for full disclosure.

---

Developed by Roberto Junior.