causal-toolkit-pmathew 0.1.2

A Python package for causal inference methods including ATE estimation, propensity score methods, and meta-learners.

causal_toolkit_pmathew

causal_toolkit_pmathew packages the causal implementations into an installable causal inference toolkit.

Package contents

• causal_toolkit_pmathew/rct.py: randomized controlled trial estimators from Week 02
• causal_toolkit_pmathew/propensity.py: inverse propensity weighting and doubly robust estimators from Week 03
• causal_toolkit_pmathew/meta_learners.py: S-, T-, X-learner, and double machine learning CATE estimators from Weeks 04 and 05
• tests/: pytest coverage for the required modules

Installation

# Install from PyPI
pip install causal_toolkit_pmathew

# Or install from source
git clone https://github.com/prinupmathew/causal_toolkit_pmathew.git
cd causal_toolkit_pmathew
uv pip install -e .
python -m uv pip install -e .

Run tests

uv run pytest
python -m uv run pytest

Run pytest module with coverage

python -m pytest tests/ -v --cov=causal_toolkit_pmathew --cov-report=term-missing

Publish to PyPI

• Create accounts:
• Generate API tokens:
  • Go to Account Settings → API tokens → Add token
  • Scope: "Entire account"
  • Save token securely (starts with pypi-)
• Publish to PyPI:

python -m pip install --upgrade twine build

# Set pypi credential locally in terminal (cmd)
set TWINE_USERNAME=__token__
set TWINE_PASSWORD=pypi-your-token-here

# Set pypi credential locally in terminal (bash)
export TWINE_USERNAME="__token__"
export TWINE_PASSWORD="pypi-your-token-here"

# Set pypi credential locally in terminal (powershell)
$env:TWINE_USERNAME="__token__"
$env:TWINE_PASSWORD="pypi-your-token-here"

# Verify tests pass
pytest tests/ -v

# Clean and rebuild
python -m build
python -m twine check dist/*
python -m twine upload dist/*

Automated publishing (GitHub Release -> PyPI)

1. Add repository secret PYPI_API_TOKEN in GitHub Actions settings.
2. Create and publish a GitHub Release (for example tag v0.1.1).
3. GitHub Actions workflow publish.yml builds and uploads to PyPI automatically.

Usage

import pandas as pd

from causal_toolkit_pmathew import (
	calculate_ate_ci,
	doubly_robust,
	ipw,
	s_learner_discrete,
	t_learner_discrete,
)

trial_data = pd.DataFrame({
	"T": [1, 1, 0, 0],
	"Y": [11.2, 10.8, 8.9, 9.1],
})

ate, ci_lower, ci_upper = calculate_ate_ci(trial_data)

observational_data = pd.DataFrame({
	"x": [0.2, 0.7, -0.1, 1.3],
	"t": [1, 1, 0, 0],
	"y": [3.4, 2.8, 1.1, 1.6],
})

ipw_estimate = ipw(observational_data, "x", "t", "y")
dr_estimate = doubly_robust(observational_data, "x", "t", "y")

train = observational_data.rename(columns={"x": "x1"}).assign(x2=[1.0, 0.5, 0.1, -0.2])
test = train.copy()

s_cate = s_learner_discrete(train, test, ["x1", "x2"], "t", "y")
t_cate = t_learner_discrete(train, test, ["x1", "x2"], "t", "y")

API

• calculate_ate_ci(data, alpha=0.05): estimate the average treatment effect and a two-sided confidence interval from an RCT dataset with T and Y columns
• calculate_ate_pvalue(data): estimate the ATE, test statistic, and two-sided p-value for an RCT dataset with T and Y columns
• ipw(df, ps_formula, T, Y): compute an inverse propensity weighted ATE
• doubly_robust(df, formula, T, Y): compute a doubly robust ATE using propensity and outcome models
• s_learner_discrete(train, test, X, T, y): estimate CATE with a single outcome model
• t_learner_discrete(train, test, X, T, y): estimate CATE with separate treated and control models
• x_learner_discrete(train, test, X, T, y): estimate CATE with the X-learner procedure
• double_ml_cate(train, test, X, T, y): estimate CATE using residualized treatment and outcome models