Optimizers¶
Apart from the one(s) below, any optimizer in tensorflow.keras.optimizers can be used.
-
class
libspn_keras.optimizers.OnlineExpectationMaximization(learning_rate=0.01, accumulate_batches=1, name='OnlineEM', **kwargs)¶ Online expectation maximization.
Requires sum layers to use any of the EM-based
SumOpBaseinstances, such asSumOpEMBackpropSumOpHardEMBackprop.- Parameters
learning_rate (
float) – Learning rate for EM. If learning rate is \(\eta\), then updates are given by: \(w \leftarrow (1-\eta)w + \eta \Delta w\)accumulate_batches (
int) – The number of batches to accumulate gradients before applying updates.name (
str) – Name of the optimizerkwargs – Remaining kwargs to pass to
Optimizersuperclass