def _grow_ensemble_not_ready_fn():
# Don't grow the ensemble, just update the stamp.
return training_ops.grow_tree_ensemble(
tree_ensemble_handle=self._ensemble_handle,
stamp_token=ensemble_stamp,
next_stamp_token=next_ensemble_stamp,
learning_rate=0,
partition_ids=[],
gains=[],
splits=[],
learner_config=self._learner_config_serialized,
dropout_seed=dropout_seed,
center_bias=self._center_bias)
def _grow_ensemble_not_ready_fn():
# Don't grow the ensemble, just update the stamp.
return training_ops.grow_tree_ensemble(
tree_ensemble_handle=self._ensemble_handle,
stamp_token=ensemble_stamp,
next_stamp_token=next_ensemble_stamp,
learning_rate=0,
partition_ids=[],
gains=[],
splits=[],
learner_config=self._learner_config_serialized,
dropout_seed=dropout_seed,
center_bias=self._center_bias)
def _grow_ensemble_ready_fn():
# Grow the ensemble given the current candidates.
sizes = array_ops.unstack(split_sizes)
partition_ids_list = list(array_ops.split(partition_ids, sizes, axis=0))
gains_list = list(array_ops.split(gains, sizes, axis=0))
split_info_list = list(array_ops.split(split_infos, sizes, axis=0))
return training_ops.grow_tree_ensemble(
tree_ensemble_handle=self._ensemble_handle,
stamp_token=ensemble_stamp,
next_stamp_token=next_ensemble_stamp,
learning_rate=learning_rate,
partition_ids=partition_ids_list,
gains=gains_list,
splits=split_info_list,
learner_config=self._learner_config_serialized,
dropout_seed=dropout_seed,
center_bias=self._center_bias)
