peegeem

Probabilistic graphical models that are fun to work with

This simple library is meant as a demo that shows what might be possible when you combine domain specific languages with bespoke UI tools. You might end up with a domain specific environment as a result!

Install

You can install via pip.

uv pip install peegeem

Usage

This is how you might use this experiment.

from peegeem import DAG

# Define the DAG for the PGM, nodes is a list of column names, edges is a list of tuples
dag = DAG(nodes, edges, dataframe)

# Get variables out
outcome, smoker, age = dag.get_variables()

# Use variables to construct a probablistic query
P(outcome | (smoker == "Yes") & (age > 40))

# Latex utility, why not?
P.to_latex(outcome | (smoker == "Yes") & (age > 40))

The goal is to have an API that really closely mimics the math notation, so stuff like this:

$$ P(\text{outcome} \mid do(\text{smoker}=\texttt{Yes}), \text{age}>40) $$

That means that we indeed also have a do function, though this needs more extensive testing.

# You can also get crazy fancy 
P(A & B | C & do(D))

Demo

For a solid demo, download and run this notebook locally.

You can run the notebook in a sandboxed environment with uv. Navigate to the nbs directory and run:

uvx marimo edit --sandbox __init__.py