Causal Inference Deep Dive

Project: projects/causalml-kaggle-showcase

Why This Deep Dive

Use this project when you need to move from prediction to intervention decisions:

• ATE and subgroup treatment effect reasoning,
• uplift ranking for targeted actions,
• Qini and uplift-at-k evaluation,
• budget-aware policy simulation,
• confounding checks and interpretability.

Quickstart

cd projects/causalml-kaggle-showcase
make sync
make download
make pipeline
make policy
make confounding
make verify

Key Terms In Practice

• ATE: average effect across all users.
• CATE / tau(x): estimated effect for user segments defined by features.
• Uplift score: model estimate used to rank who to treat first under budget constraints.

Example: Connect Tau To Targeting

If a model predicts:

• User A: tau(x)=0.07
• User B: tau(x)=0.01

and budget allows one contact, policy should prioritize user A because expected incremental impact is higher.

Example: Policy Run

cd projects/causalml-kaggle-showcase
python scripts/policy_simulator.py

Expected outputs include:

• artifacts/policy_simulation.csv
• artifacts/policy_recommendations.csv
• artifacts/figures/policy_incremental_conversions.png

Example: Read Top Policy Rows

cd projects/causalml-kaggle-showcase
python - <<'PY'
import pandas as pd
df = pd.read_csv("artifacts/policy_recommendations.csv")
print(df.head(8).to_string(index=False))
PY

How To Interpret Outputs

1. Positive ATE does not imply everyone benefits; inspect subgroup uplift.
2. Favor models that improve uplift-at-k and Qini, not only raw predictive metrics.
3. If propensity overlap is weak, treat causal claims as lower confidence.
4. Policy recommendations should include budget sensitivity, not one fixed threshold.

Next Step

Cross-check production readiness with MLOps Drift Deep Dive, especially monitoring and retraining policy design.