The name pqp
is short for Pourquoi pas?. This phrase is French for why not?, because "Why not?" was the question we asked ourselves when we found there was no maintained, open-source package for structural causal modeling in Python. The package provides a convenient interface for causal modeling along with routines for identification, estimation, and visualization.
The package can be installed from PyPi using pip
:
pip install pqp
from pqp.graph import Graph
from pqp.variable import make_vars
# create variables
x, y, z = make_vars("xyz")
# the backdoor model
g = Graph([
x <= z,
y <= z,
y <= x,
])
# identification
causal_estimand = ATE(y, {x: 1}, {x: 0})
estimator = g.identify(causal_estimand)
print(estimator)
# >>> E_(y) [ Σ_(z) [ [P(x = 1, y, z) * P(z) / P(x = 1, z)] ] ] - E_(y) [ Σ_(z) [ [P(x = 0, y, z) * P(z) / P(x = 0, z)] ] ]
For more information, see the documentation at https://leo-ware.github.io/pqp/.
The source code is available at https://github.com/leo-ware/pqp.
This package was created by Leo Ware as part of his undergraduate capstone project at Minerva University.