causal-toolkit-mkhan 0.1.0

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

Causal Toolkit

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

Installation

git clone https://github.com/mokhan95/causal-toolkit-mkhan.git
cd causal-toolkit-mkhan
uv pip install -e .

Usage

RCT Analysis

from causal_toolkit_mkhan import calculate_ate_ci, calculate_ate_pvalue

ate, ci_lower, ci_upper = calculate_ate_ci(data)
ate, t_stat, p_value = calculate_ate_pvalue(data)

Propensity Score Methods

from causal_toolkit_mkhan import ipw, doubly_robust

ate_ipw = ipw(df, ps_formula="X1 + X2", T="treatment", Y="outcome")
ate_dr = doubly_robust(df, formula="X1 + X2", T="treatment", Y="outcome")

Meta-Learners

from causal_toolkit_mkhan import s_learner_discrete, t_learner_discrete, x_learner_discrete

result = s_learner_discrete(train, test, X=['x1', 'x2'], T='treatment', y='outcome')
result = t_learner_discrete(train, test, X=['x1', 'x2'], T='treatment', y='outcome')
result = x_learner_discrete(train, test, X=['x1', 'x2'], T='treatment', y='outcome')

Double ML

from causal_toolkit_mkhan import double_ml_cate

result = double_ml_cate(train, test, X=['x1', 'x2'], T='treatment', y='outcome')

API Reference

calculate_ate_ci(data, alpha=0.05)

Returns (ate, ci_lower, ci_upper) using z-test.

calculate_ate_pvalue(data)

Returns (ate, t_statistic, p_value).

ipw(df, ps_formula, T, Y)

Inverse Propensity Weighting estimation.

doubly_robust(df, formula, T, Y)

Doubly Robust estimation.

s_learner_discrete(train, test, X, T, y)

S-learner for discrete treatment CATE estimation.

t_learner_discrete(train, test, X, T, y)

T-learner for discrete treatment CATE estimation.

x_learner_discrete(train, test, X, T, y)

X-learner for discrete treatment CATE estimation.

double_ml_cate(train, test, X, T, y)

Double ML for continuous treatment CATE estimation.

Testing

uv run pytest tests/ -v

Dependencies

• pandas >= 1.3.0
• numpy >= 1.21.0
• scipy >= 1.7.0
• scikit-learn >= 1.0.0
• lightgbm >= 3.3.0
• patsy >= 0.5.0

License

MIT License

Author

mkhan ([e-mail])