pgmpy: Probabilistic Graphical Models

1.1.0 · active · verified Thu Apr 16

pgmpy is a Python library for working with Probabilistic Graphical Models (PGMs). It provides implementations of various models, inference algorithms, and learning algorithms for Bayesian Networks, Markov Networks, and other causal and probabilistic reasoning tasks. The current version is 1.1.0, with minor releases occurring periodically and major versions like 1.0.0 introducing significant breaking changes.

Common errors

```
AttributeError: module 'pgmpy.models' has no attribute 'BayesianModel'
```
cause Attempting to import or use the `BayesianModel` class in pgmpy versions 1.0.0 or later.

fix The `BayesianModel` class was renamed to `BayesianNetwork`. Update your code to `from pgmpy.models import BayesianNetwork`.
```
ValueError: The sum of the probability values for Variable [variable_name] is not 1.
```
cause When defining a `TabularCPD`, the probabilities for each state combination of the parent variables must sum to 1. This error indicates a mathematical inconsistency in your CPD definition.

fix Review the `values` array in your `TabularCPD` definition. Ensure that each column (representing a parent configuration) sums to 1. Double-check the order of variables and their cardinalities.
```
ImportError: cannot import name 'VariableElimination' from 'pgmpy.inference'
```
cause The module path for exact inference algorithms changed. `VariableElimination` is no longer directly under `pgmpy.inference`.

fix Import `VariableElimination` from its specific submodule: `from pgmpy.inference.exact import VariableElimination`.

Warnings

breaking Major breaking changes were introduced in v1.0.0. Specifically, `BayesianModel` was renamed to `BayesianNetwork` and `MarkovModel` to `MarkovNetwork`. Code using the old class names will fail with an `AttributeError`.
Fix: Update class imports from `BayesianModel` to `BayesianNetwork` and `MarkovModel` to `MarkovNetwork`.
gotcha The order of evidence variables and their cardinality is critical when defining `TabularCPD`s. An incorrect order or mismatch between `evidence` and `evidence_card` will lead to incorrect probability distributions or errors.
Fix: Carefully align the `values` matrix with the `evidence` variables' order and `evidence_card`. Consult the documentation for the specific indexing order (e.g., last variable changes fastest).
gotcha Inference on large or densely connected graphical models can be computationally expensive and memory intensive, especially for exact inference methods like Variable Elimination. Consider approximate inference for such cases.
Fix: For large models, explore approximate inference algorithms (e.g., `pgmpy.inference.sampling`) or simplify the model structure.

Install

pip install pgmpy Install stable version

Imports

BayesianNetwork
wrong:
```
from pgmpy.models import BayesianModel
```
correct:
```
from pgmpy.models import BayesianNetwork
```
The class was renamed from `BayesianModel` to `BayesianNetwork` in v1.0.0.
MarkovNetwork
wrong:
```
from pgmpy.models import MarkovModel
```
correct:
```
from pgmpy.models import MarkovNetwork
```
The class was renamed from `MarkovModel` to `MarkovNetwork` in v1.0.0.
TabularCPD
```
from pgmpy.factors.discrete import TabularCPD
```
Used for defining Conditional Probability Distributions for discrete variables.
VariableElimination
wrong:
```
from pgmpy.inference import VariableElimination
```
correct:
```
from pgmpy.inference.exact import VariableElimination
```
Exact inference algorithms are now explicitly under `pgmpy.inference.exact`.

Quickstart

This quickstart demonstrates how to define a Bayesian Network, add Conditional Probability Distributions (CPDs) to it, check its validity, and perform exact inference using the Variable Elimination algorithm. The example builds a small network for a student's performance.

from pgmpy.models import BayesianNetwork
from pgmpy.factors.discrete import TabularCPD
from pgmpy.inference import VariableElimination

# 1. Define the network structure
# D: Difficulty (Easy, Hard), I: Intelligence (Low, High)
# G: Grade (A, B, C), L: Letter (Good, Bad), S: SAT (Low, High)
model = BayesianNetwork([('D', 'G'), ('I', 'G'), ('G', 'L'), ('I', 'S')])

# 2. Define Conditional Probability Distributions (CPDs)
cpd_d = TabularCPD(variable='D', variable_card=2, values=[[0.6], [0.4]])
cpd_i = TabularCPD(variable='I', variable_card=2, values=[[0.7], [0.3]])

# G depends on I and D. Order of evidence: I, D
cpd_g = TabularCPD(variable='G', variable_card=3,
                   values=[[0.3, 0.05, 0.9, 0.5],
                           [0.4, 0.25, 0.08, 0.3],
                           [0.3, 0.7, 0.02, 0.2]],
                   evidence=['I', 'D'], evidence_card=[2, 2])

cpd_l = TabularCPD(variable='L', variable_card=2,
                   values=[[0.1, 0.4, 0.99],
                           [0.9, 0.6, 0.01]],
                   evidence=['G'], evidence_card=[3])

cpd_s = TabularCPD(variable='S', variable_card=2,
                   values=[[0.95, 0.2],
                           [0.05, 0.8]],
                   evidence=['I'], evidence_card=[2])

# 3. Add CPDs to the model
model.add_cpds(cpd_d, cpd_i, cpd_g, cpd_l, cpd_s)

# 4. Check if the model is valid (optional but good practice)
assert model.check_model(), "Model is not valid!"

# 5. Perform inference
inference = VariableElimination(model)

# Query for P(L | D=0, I=1)
result = inference.query(variables=['L'], evidence={'D': 0, 'I': 1})
print("P(L | D=0, I=1):")
print(result)

# Query for P(G | S=0)
result_g_s = inference.query(variables=['G'], evidence={'S': 0})
print("\nP(G | S=0):")
print(result_g_s)

view raw JSON →