update sequential reco models: add serving example

[Leavingseason]

Description

Update deeprec's sequential recommender packages. Provide an example for users how can we use the trained model for serving purpose.

Related Issues

#1233
#1068

Checklist:

• I have followed the contribution guidelines and code style for this project.
• I have added tests covering my contributions.
• I have updated the documentation accordingly.
• This PR is being made to staging and not master.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[miguelgfierro]

hey @Leavingseason this is awesome, but the notebook is not loading for me, I think it might be because it has too many logs.

Do you think we can clean or reduce the output of the cell:

G = tf.Graph()
with tf.gfile.GFile(
        os.path.join(hparams.MODEL_DIR, "serving_model.pb"),
        'rb'
) as f, G.as_default():
    graph_def_optimized = tf.GraphDef()
    graph_def_optimized.ParseFromString(f.read())
    print('graph_def_optimized = ' + str(graph_def_optimized))


with tf.Session(graph=G) as sess:
    tf.import_graph_def(graph_def_optimized)

    model = LoadFrozedPredModel(sess.graph)
    
    serving_output_file = os.path.join(data_path, r'output_serving.txt')  
    iterator = input_creator(hparams, tf.Graph())
    infer_as_serving(model, test_file, serving_output_file, hparams, iterator, sess)

[Leavingseason]

hey @Leavingseason this is awesome, but the notebook is not loading for me, I think it might be because it has too many logs.

Do you think we can clean or reduce the output of the cell:

G = tf.Graph()
with tf.gfile.GFile(
        os.path.join(hparams.MODEL_DIR, "serving_model.pb"),
        'rb'
) as f, G.as_default():
    graph_def_optimized = tf.GraphDef()
    graph_def_optimized.ParseFromString(f.read())
    print('graph_def_optimized = ' + str(graph_def_optimized))


with tf.Session(graph=G) as sess:
    tf.import_graph_def(graph_def_optimized)

    model = LoadFrozedPredModel(sess.graph)
    
    serving_output_file = os.path.join(data_path, r'output_serving.txt')  
    iterator = input_creator(hparams, tf.Graph())
    infer_as_serving(model, test_file, serving_output_file, hparams, iterator, sess)

Sure. I have removed the printing of graph content

[miguelgfierro]

Awesome!

[aidenpearce001]

Why i get error AssertionError: pred is not in graph, when running this code

[Leavingseason]

@aidenpearce001 are you using the latest code? In this PR, I update the base model with one line of code "pred = tf.identity(pred, name='pred')"

[aidenpearce001]

@aidenpearce001 are you using the latest code? In this PR, I update the base model with one line of code "pred = tf.identity(pred, name='pred')"

Thanks you for helping me. I want to ask about model output when using model.predict(), its just return a number but i want it to return probability for each items so i can recommend a list of item for user. How can i do it?

[Leavingseason]

Each line is an instance of <user, item> pair. So enumerate all the items you need to make prediction for the user. E.g., if you have 100 items to rate, you need to generate 100 lines of <user, item> pairs. SLi-Rec model is for ranking purpose, not suitable for item retrieval.