TensorFlow Estimator to TensorFlow JS

Question

With the help of the Stack Overflow community, I have started to understand the complexity of TensorFlow's "high level" API, Estimator. I, perhaps foolishly, did so thinking that if TensorFlow and TensorFlow JS were to get along, most likely it would be through the Estimator API...

So in this Colab, I have a simple custom Estimator which is simply wired up so that methods like .train_and_evaluate work and after "training" the Estimator is exported via export_savedmodel.

Now, suppose I want to go ahead and use this trained model in the browser via TensorFlow js.

To my luck, there is a guide on how to convert a saved_model for TensorFlow JS:

tensorflowjs_converter \
    --input_format=tf_saved_model \
    --output_node_names='MobilenetV1/Predictions/Reshape_1' \
    --saved_model_tags=serve \
    /mobilenet/saved_model \
    /mobilenet/web_model

While the arguments have a description...

• --output_node_names The names of the output nodes, separated by commas.
• --saved_model_tags Only applicable to SavedModel conversion, Tags of the MetaGraphDef to load, in comma separated format. Defaults to serve.
• --signature_name Only applicable to TensorFlow Hub module conversion, signature to load. Defaults to default. See https://www.tensorflow.org/hub/common_signatures/.

I am not sure what I am supposed to replace these with for the demo estimator found in the Colab.

Why? Well for starters, the best_model exporter, which uses the serving_input_receiver_fn, has different output when loaded via

from tensorflow.contrib import predictor
predict_fn = predictor.from_saved_model('<exported_location>')

than from

estimator.predict(lambda: predict_input_fn(pred_features), yield_single_examples=False)

namely, the key of the predicted features features outputted in the former is "outputs" and "labels" in the latter.

which, forgive my mini-rant, why doesn't Estimator have a way to load an exported model?

so:

1. what are my "output_node_names"?

2. since this is a SavedModel, what tags do I need?

I would greatly appreciate any guidance.