Causal Inference Toolkit¶

Production-ready causal inference for data scientists and ML engineers.

Why This Toolkit?¶

Problem	Solution
"DoWhy is great but I need EconML's CATE estimators too"	Unified wrapper — one API for identification, estimation, refutation
"Sensitivity analysis is an afterthought"	First-class citizens — Rosenbaum, Cinelli-Hazlett, E-values, TIPS curves built in
"A/B testing needs both frequentist and Bayesian"	Complete ABTestAnalyzer — z-test, t-test, SPRT, mSPRT, Beta-Binomial, ROPE, expected loss
"Uplift modeling is scattered across libraries"	UpliftModeler — T/S/X/R/DR-learners + Causal Forest + Qini/AUUC evaluation
"I need to show my work"	Publication-ready viz — DAGs with backdoor highlighting, forest plots, love plots, Qini curves

Quick Links¶


🚀 Quickstart	⚙️ Installation	📝 Configuration
🧠 Core Concepts	🔧 DoWhy Wrapper	🌲 EconML Wrapper
📊 A/B Testing	🎯 Uplift Modeling	📈 Visualization
🔍 Sensitivity Analysis	💻 CLI Reference	📚 API Reference

Example: Full Pipeline in 10 Lines¶

from causal_toolkit import CausalModel, DoWhyWrapper, SensitivityAnalyzer
from causal_toolkit.core.base import EstimatorType, IdentificationStrategy

# 1. Load data
df = pd.read_csv("data.csv")

# 2. Define model
model = CausalModel(df, treatment="treatment", outcome="outcome", 
                    common_causes=["age", "income", "education"])

# 3. Identify + Estimate
dowhy = DoWhyWrapper(model)
estimand = dowhy.identify(IdentificationStrategy.BACKDOOR)
estimate = dowhy.estimate(EstimatorType.DOUBLY_ROBUST)

# 4. Refute
for r in dowhy.refute():
    print(r)

# 5. Sensitivity
sa = SensitivityAnalyzer(model)
sa.cinelli_hazlett(estimate)
print(sa.summarize())

Installation¶

pip install causal-toolkit
# With extras:
pip install "causal-toolkit[dev,viz,notebooks]"

Built on Giants¶

Library	Role
DoWhy	Causal identification, refutation, do-calculus
EconML	Heterogeneous treatment effects (CATE), metalearners, forests
CausalInference	Propensity score methods, matching
Statsmodels	Power analysis, proportion tests
NetworkX/Graphviz	Causal graph algorithms & rendering

License¶

MIT License — see LICENSE

Citation¶

@software{causal_toolkit,
  author = {Anuj Singh},
  title = {Causal Inference Toolkit},
  year = {2025},
  url = {https://github.com/anujsingh-cse/causal-inference-toolkit}
}