adanet.replay

Defines mechanisms for deterministically replaying an AdaNet model search.

Config

class adanet.replay.Config(best_ensemble_indices=None)[source]

Defines how to deterministically replay an AdaNet model search.

Specifically, it reconstructs the previous model and trains its components in the correct order without performing any search.

Parameters:best_ensemble_indices – A list of the best ensemble indices (one per iteration).
Returns:An adanet.replay.Config instance.
best_ensemble_indices

The best ensemble indices per iteration.

get_best_ensemble_index(iteration_number)[source]

Returns the best ensemble index given an iteration number.