pycausal-explorer 0.2.0

Python causal inference modules

Pycausal-explorer

Pycausal-explorer is a python module for causal inference and treatment effect estimation. It implements a set of algorithms that supports causal analysis.

Installation Guide

You can install the package through pip:

pip install pycausal-explorer

Basic Usage

All models are inherited from BaseCausalModel, that inherits from scikit-learn BaseEstimator. It uses scikit-learn framework to fit and predict the outcome. It implements predict_ite and predict_ate methods that return the individual treatment effect and the average treatment effect, respectively.

from pycausal_explorer.datasets.synthetic import create_synthetic_data
from pycausal_explorer.meta import XLearner

x, treatment, y = create_synthetic_data()
model = XLearner()
model.fit(x, treatment, y)
treatment_effect = model.predict_ite(x)

Current Implemented Models

This version currently implements propensity score and iptw in the reweight package, linear regression in the linear package, causal forests in forest package and x-learn in meta package.

Using Pipelines

Pycausal-explorer has a Pipeline class inherited from scikit-learn Pipeline. It implements the method predict_ite, so it can be used pro predict treatment effect in a pipeline:

from sklearn.preprocessing import StandardScaler

from pycausal_explorer.datasets.synthetic import create_synthetic_data
from pycausal_explorer.pipeline import Pipeline
from pycausal_explorer.reweight import IPTW

x, w, y = create_synthetic_data()
pipe = Pipeline([("norm", StandardScaler()), ("clf", IPTW())])
pipe.fit(x, y, clf__treatment=w)
treatment_effect = pipe.predict_ite(x)