fc1c38580265de55e36ae30f236739d898cde1bc,census/keras/trainer/model.py,,to_savedmodel,#Any#Any#,82

Before Change

def to_savedmodel(model, export_path):
  Convert the Keras HDF5 model into TensorFlow SavedModel.
  with K.get_session() as sess:
    saved_model_util.simple_save(
        sess,
        export_path,
        inputs={"input": model.inputs[0]},
        outputs={"income": model.outputs[0]})


def to_numeric_features(features):
  Convert the pandas input features to numeric values.

After Change

  builder = saved_model_builder.SavedModelBuilder(export_path)

  signature = predict_signature_def(inputs={"input": model.inputs[0]},
                                    outputs={"income": model.outputs[0]})

  with K.get_session() as sess:
    builder.add_meta_graph_and_variables(
        sess=sess,
        tags=[tag_constants.SERVING],
        signature_def_map={
            signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY: signature}
    )
    builder.save()


def to_numeric_features(features):
  Convert the pandas input features to numeric values.

Instances