Relationship between tensor keeper, exporter and model conservation

Question

Question :

• Tensorflow Saver, Exporter, SavedModelBuilder can be used to save models. According to https://stackoverflow.com/questions/41740101/tensorflow-difference-between-saving-model-via-exporter-and-tf-train-write-graph and tensor flow , I understand that Saver is used to save training control points, while Exporter and SavedModelBuilder are used for maintenance.

  However, I do not know the differences in their results. Are variables .data - ??? - of - ??? and variables. index files generated by SavedModelBuilder are the same as cpkt-xxx.index and cpkt-xxx.data - ??? - of - ??? created by Saver?

• I still feel embarrassed about the value of tensor flow model files. I read http://cv-tricks.com/tensorflow-tutorial/save-restore-tensorflow-models-quick-complete-tutorial/ and Tensorflow: how to save / restore the model? that makes me feel more embarrassed.

  There are 4 files in the model directory:

  • graph.pbtxt
  • model.ckpt-number.data-00000-of-00001
  • model.ckpt-number.meta li>
  • model.ckpt-number.index

  Files 2 and 4 store the weights of the variables. File 3 stores the graph. Then what is 1 store?

• How to Convert Saver Outputs to SavedModelBuilder. I have a checkpoint directory and you want to export the model for maintenance. According to https://github.com/tensorflow/tensorflow/tree/master/tensorflow/python/saved_model

he should be like that

export_dir = ... ... builder = tf.saved_model.builder.SavedModelBuilder(export_dir) with tf.Session(graph=tf.Graph()) as sess: ... builder.add_meta_graph_and_variables(sess, [tf.saved_model.tag_constants.TRAINING], signature_def_map=foo_signatures, assets_collection=foo_assets) ... with tf.Session(graph=tf.Graph()) as sess: ... builder.add_meta_graph(["bar-tag", "baz-tag"]) ... builder.save() 

So, first I need to load breakpoints with:

 saver = tf.train.import_meta_graph('model-number.meta') saver.restore(sess, tf.train.latest_checkpoint('./')) 

And then use this sess for the builder.

I'm right?

Answer 1

SavedModel is a service format created using SavedModelBuilder. Best practice is for your tutorial code to call SavedModelBuilder and feed the resulting TF-Serving output files. If you do this, you do not need to understand the details of which files are created :)

The document [1] talks about the structure of files inside the SavedModel directory.

[1] https://www.tensorflow.org/programmers_guide/saved_model