spn_moments.py

import numpy as np

import deeprob.spn.structure as spn
import deeprob.spn.algorithms as spnalg

if __name__ == '__main__':
    # Instantiate a simple SPN
    root = spn.Sum(children=[
        spn.Product(children=[
            spn.Gaussian(0, 0.5, 1.0),
            spn.Gaussian(1, -1.0, 0.5),
            spn.Gaussian(2, 0.0, 1.0)
        ]),
        spn.Product(children=[
            spn.Gaussian(0, 0.5, 1.5),
            spn.Gaussian(1, 0.0, 0.2),
            spn.Gaussian(2, 2.0, 1.0)
        ])
    ], weights=[0.8, 0.2])

    # Initialize the IDs of the SPN nodes
    spn.assign_ids(root)

    # Compute and print four moments w.r.t. to each random variable
    expectation = spnalg.expectation(root)
    variance = spnalg.variance(root)
    skewness = spnalg.skewness(root)
    kurtosis = spnalg.kurtosis(root)
    print("Expectation: {}".format(np.round(expectation, 2)))
    print("Variance: {}".format(np.round(variance, 2)))
    print("Skewness: {}".format(np.round(skewness, 2)))
    print("Kurtosis: {}".format(np.round(kurtosis, 2)))