Kind request to tensorflow model developers

First, I should say that I'm hugely indebted that you are releasing your code and/or trained models. A million thanks!

You would do me (and potentially others) a huge favour if you were to make a few tiny tweaks to your code, and save me from forking so many repos just to make these tiny changes :) (such as this, this or this).

Details are explained here, but in short it would be amazing if you could:

1. Give input / output operators human readable names, so that feeding and fetching can be done by name (this of course includes ops like LSTM states, and may involve adding stacks or tf.identity if needed e.g. see this commit). Heck why not throw in a comment or two indicating what these names are! :)

2. If you have an export option, why not throw in a pruned and frozen graph_def export? AFAIK it's just one line of code ;)

    tf.train.write_graph(
            tf.graph_util.convert_variables_to_constants(sess, sess.graph_def, [output nodes])
            output_dir,
            'graph_frz.pb',
            as_text=False)
   

Thank you please!