View source on GitHub

Function builder for a dnn logit_fn.

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.