SPPL (Sum-Product Probabilistic Language)

2.0.4 verified Sat May 09 auth: no python

SPPL is a probabilistic programming language for specifying and manipulating sum-product expressions. Version 2.0.4 targets Python >=3.8. The library enables concise modeling of discrete and continuous random variables with automatic inference.

pip install sppl

Common errors

error ImportError: cannot import name 'SPPLDistribution' from 'sppl' ↓

cause Trying to import SPPLDistribution from the top-level sppl package instead of the submodule.

fix

Use 'from sppl.distributions import SPPLDistribution'

error AttributeError: 'Variable' object has no attribute 'distribution' ↓

cause Accessing .distribution attribute that was renamed or restructured.

fix

Use the new API: see the documentation for accessing the distribution from a Variable.

error TypeError: __init__() missing 1 required positional argument: 'dist' ↓

cause Creating Variable without providing the distribution argument correctly.

fix

Ensure you pass the distribution as the second argument to Variable: Variable('name', distribution).

Warnings

breaking In version 2.0.0, the package structure changed: distributions are now in 'sppl.distributions' submodule, not directly under 'sppl'. ↓

fix Update imports: use 'from sppl.distributions import SPPLDistribution' instead of 'from sppl import SPPLDistribution'.

deprecated The function 'sppl.distributions.Distribution' is deprecated in favor of 'SPPLDistribution'. ↓

fix Replace 'Distribution' with 'SPPLDistribution' in your code.

gotcha Model variables must be defined inside a 'with model:' block, otherwise they are not added to the model. ↓

fix Always use 'with model:' context manager when defining variables.

Imports

SPPLDistribution

wrong

from sppl import SPPLDistribution

correct

from sppl.distributions import SPPLDistribution

SPPLDistribution is in the distributions submodule.

Model
```
from sppl import Model
```
Direct import from the top-level package.
Variable
```
from sppl import Variable
```
Direct import from the top-level package.

Quickstart

Creates a simple probabilistic model with a Bernoulli variable.

from sppl import Model, Variable
from sppl.distributions import SPPLDistribution
import os

# Define a simple model: flip a fair coin
model = Model()
with model:
    x = Variable('x', SPPLDistribution.bernoulli(0.5))

# Check the model structure
print(model)