gbdrt / mu-ppl

µ-PPL

mu-PPL is a micro probabilistic programming language (PPL) on top of Python (mu is for the greek letter µ). Following recent PPLs (WebPPL, Pyro, Gen.jl) mu-PPL provides the following probabilistic operators:

• sample to draw a sample from a distribution
• assume to condition the model on a boolean condition (hard conditionning)
• factor / observe to update the score of the current execution (soft conditionning)

Getting started

mu-PPL can be installed locally with:

$ pip install git+https://github.com/gbdrt/mu-ppl

Here is the mu-PPL program that computes the bias of coin from a series of observations.

from typing import List
from mu_ppl import *

def coin(obs: List[int]) -> float:
    p = sample(Uniform(0, 1), name="p")
    for o in obs:
        observe(Bernoulli(p), o)
    return p

with ImportanceSampling(num_particles=10000):
    dist: Categorical[float] = infer(coin, [0, 0, 0, 0, 0, 0, 0, 0, 1, 1])  # type: ignore
    print(dist.stats())

You should get the following output:

100%|███████████████████████████████████| 10000/10000 [00:02<00:00, 3785.50it/s]
(0.2511526566307631, 0.11951005324220002)

More examples are available in the examples directory.

Notebooks

The following introductory notebooks can be run locally or in Google Colab.

• Introduction
• Bayesian Reasoning

EJCIM 2024

• An introduction to probabilistic programming with mu-PPL

• Introduction to Sequential Monte Carlo methods (SMC)