tf.contrib.kfac.estimator.make_fisher_estimator

tf.contrib.kfac.estimator.make_fisher_estimator(
    placement_strategy=None,
    **kwargs
)

Defined in tensorflow/contrib/kfac/python/ops/estimator.py.

Creates Fisher estimator instances based on the placement strategy.

For example if the placement_strategy is 'round_robin' then FisherEstimatorRoundRobin instance is returned.

Args:

  • placement_strategy: string, Strategy to be used for placing covariance variables, covariance ops and inverse ops. Check placement.FisherEstimatorRoundRobin for a concrete example. **kwargs: Arguments to be passed into FisherEstimator class initializer.

Returns:

An instance of class which inherits from FisherEstimator and the mixin which implements specific placement strategy. See, FisherEstimatorRoundRobin which inherits from FisherEstimator and RoundRobinPlacementMixin.

Raises:

  • ValueError: If the placement_strategy is not equal to 'round_robin'.