Tutorial notebooks for Linear Gaussian Bayesian Networks

[ankurankan]

We currently have an example notebook: https://pgmpy.org/examples/Creating%20a%20Discrete%20Bayesian%20Network.html for creating discrete Bayesian networks but not for Linear Gaussian BNs. Write a similar notebook with tutorial on creating gaussian bayesian networks.

[Vishnu-sai-teja]

Hey @ankurankan, can you assign this to me ?

[guptaabhishek07]

Importing Libraries

from pgmpy.models import BayesianNetwork from pgmpy.factors.continuous import GaussianCPD

Creating a Bayesian Network

model = BayesianNetwork([('Age', 'BloodPressure'), ('Smoking', 'BloodPressure'), ('Age', 'HeartDisease'), ('BloodPressure', 'HeartDisease')])

Adding CPDs

cpd_age = GaussianCPD('Age', mu=30, sigma=5) cpd_smoking = GaussianCPD('Smoking', mu=0.2, sigma=0.1) cpd_bp = GaussianCPD('BloodPressure', parents=['Age', 'Smoking'], mu=lambda pa: 110 + pa['Age'] + 20 * pa['Smoking'], sigma=15) cpd_hd = GaussianCPD('HeartDisease', parents=['Age', 'BloodPressure'], mu=lambda pa: 1/(1 + np.exp(-(pa['Age'] - 50)/10 - pa['BloodPressure']/20)), sigma=0.1)

model.add_cpds(cpd_age, cpd_smoking, cpd_bp, cpd_hd)

Inference

print(model.query(variables=['HeartDisease'], evidence={'Age': 40, 'Smoking': 0}))

This code defines a simple Gaussian Bayesian network with four variables: Age, Smoking, BloodPressure, and HeartDisease. The CPDs are defined using the GaussianCPD class, which specifies the mean and sigma of the Gaussian distribution for each variable. The parents argument is used to specify the parent variables for each CPD.

The query function is then used to perform inference on the network. In this case, we are querying for the probability of HeartDisease given that Age is 40 and Smoking is 0.