diff options
Diffstat (limited to 'tensorflow/contrib/estimator/python/estimator/dnn_linear_combined.py')
| -rw-r--r-- | tensorflow/contrib/estimator/python/estimator/dnn_linear_combined.py | 11 |
1 files changed, 9 insertions, 2 deletions
diff --git a/tensorflow/contrib/estimator/python/estimator/dnn_linear_combined.py b/tensorflow/contrib/estimator/python/estimator/dnn_linear_combined.py index 894a295498..2eef60c39f 100644 --- a/tensorflow/contrib/estimator/python/estimator/dnn_linear_combined.py +++ b/tensorflow/contrib/estimator/python/estimator/dnn_linear_combined.py @@ -110,7 +110,8 @@ class DNNLinearCombinedEstimator(estimator.Estimator): dnn_activation_fn=nn.relu, dnn_dropout=None, input_layer_partitioner=None, - config=None): + config=None, + linear_sparse_combiner='sum'): """Initializes a DNNLinearCombinedEstimator instance. Args: @@ -142,6 +143,11 @@ class DNNLinearCombinedEstimator(estimator.Estimator): input_layer_partitioner: Partitioner for input layer. Defaults to `min_max_variable_partitioner` with `min_slice_size` 64 << 20. config: RunConfig object to configure the runtime settings. + linear_sparse_combiner: A string specifying how to reduce the linear model + if a categorical column is multivalent. One of "mean", "sqrtn", and + "sum" -- these are effectively different ways to do example-level + normalization, which can be useful for bag-of-words features. For more + details, see @{tf.feature_column.linear_model$linear_model}. Raises: ValueError: If both linear_feature_columns and dnn_features_columns are @@ -169,7 +175,8 @@ class DNNLinearCombinedEstimator(estimator.Estimator): dnn_activation_fn=dnn_activation_fn, dnn_dropout=dnn_dropout, input_layer_partitioner=input_layer_partitioner, - config=config) + config=config, + linear_sparse_combiner=linear_sparse_combiner) super(DNNLinearCombinedEstimator, self).__init__( model_fn=_model_fn, model_dir=model_dir, config=config) |