Trains models and prints debug info.

config dictionary of test case parameters. See tests for TensorFlow Lattice layers.
training_data tuple: (training_inputs, labels) where training_inputs and labels are proper data to train models passed via other parameters.
keras_model Keras model to train on training_data.
input_dtype dtype for input conversion.
label_dtype dtype for label conversion.

Loss measured on training data and tf.session() if one was initialized explicitly during training.