Aggregated posterior responsability for each component
Sum of posterior cluster responsibilities
Sum of weighted outer products
Aggregated log-likelihood
Batch size counter
Batch size counter
Number of components in the model
Aggregated log-likelihood
Adder for other MetricAggregator instances
Adder for other MetricAggregator instances
Used for further aggregation between each worker's aggregates
Sum of weighted outer products
Sum of posterior cluster responsibilities
Aggregated posterior responsability for each component
Distributed aggregator of necessary statistics
In each worker it computes and aggregates the current batch log-likelihood, and the terms that will later be used to compute the gradient. The class structure is based on Spark's ExpectationSum