View source on GitHub

Function builder for a dnn logit_fn.

    units, hidden_units, feature_columns, activation_fn, dropout,
    input_layer_partitioner, batch_norm


  • units: An int indicating the dimension of the logit layer. In the MultiHead case, this should be the sum of all component Heads' logit dimensions.
  • hidden_units: Iterable of integer number of hidden units per layer.
  • feature_columns: Iterable of feature_column._FeatureColumn model inputs.
  • activation_fn: Activation function applied to each layer.
  • dropout: When not None, the probability we will drop out a given coordinate.
  • input_layer_partitioner: Partitioner for input layer.
  • batch_norm: Whether to use batch normalization after each hidden layer.


A logit_fn (see below).


  • ValueError: If units is not an int.