How to save model in ML Engine to train?

I am trying to upload a model to ML Engine to train it once it is deployed here, what I want is to be able to train it and then do perditions, all from ML Engine.

I am using Python as a language, I have already managed to create my model with Tensorflow and train it locally, then I deploy it in ML Engine and from there I have been able to make permissions. But I can not upload it and train it from there.

    signature = tf.saved_model.signature_def_utils.predict_signature_def(
inputs={signature_constants.PREDICT_INPUTS: x},
outputs={signature_constants.PREDICT_OUTPUTS: pred})




builder.add_meta_graph_and_variables(
        sess=sess,
        tags=[tag_constants.SERVING],
        clear_devices = True,
        signature_def_map={
            signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY: signature}
    )

This is the part of code that I have to deploy it once trained. How should I change to be trained from ML Engine and then make my perditions?