Optimizers

Apart from the one(s) below, any optimizer in tensorflow.keras.optimizers can be used.

class libspn_keras.optimizers.OnlineExpectationMaximization(learning_rate=0.05, gliding_average=True, **kwargs)

Online expectation maximization which requires sum layers to use any of the EM-based SumOpBase instances.

Internally, this is just an SGD optimizer with unit learning rate.