Expectation maximization iteratively produces an estimate of sufficient statistics for learnable parameters, then maximizes the parameters according to the estimate.
Methods for creating probabilistic factors associated with elements and their sufficient statistics.
Expectation maximization iteratively produces an estimate of sufficient statistics for learnable parameters, then maximizes the parameters according to the estimate. It uses an factored inference algorithm, SufficientStatisticsVariableElimination, to produce the estimate of the sufficient statistics. This class can be extended with a different expectation or maximization algorithm; see the code for details.