SavedModel_test.py

### We will try to seralize and desearlaize a graph that is using the new `get_single_element` function of the Dataset API
### You will see that it does not desearlize gracefully.


#### Part 1: Build arbitrary graph using Dataset API and new get_single_element function

import numpy as np
import tensorflow as tf
from tensorflow.contrib.data import Dataset, Iterator

BATCH_SIZE=100
example = tf.placeholder(tf.float32, shape=[None, 5], name="input_example") # 1 example has 5 features

w = tf.Variable(tf.random_normal([1, 5], stddev=0.35),name="weights")
b = tf.Variable(tf.zeros([5]), name="biases")

def preprocessing_fn(_in):
  return _in+0.0

dataset = (tf.data.Dataset.from_tensor_slices(example)
           .map(preprocessing_fn, num_parallel_calls=BATCH_SIZE)
           .batch(BATCH_SIZE))

x = tf.contrib.data.get_single_element(dataset) #new per fb180d5 ~20 days ago

z = tf.sigmoid(tf.multiply(w,x)+b)

init_op = tf.global_variables_initializer()


### Part 2: Export the Graph using SavedModel API

with tf.Session() as sess: 
    sess.run(init_op)
    print sess.run(z, feed_dict={example:np.random.random((2,5))})
    
    #export
    from datetime import datetime
    from tensorflow.python.saved_model import builder
    import os

    version = int(datetime.now().strftime("%s"))
    export_path = os.path.join(tf.compat.as_bytes("_versions"), tf.compat.as_bytes(str(version)))
    print 'Exporting trained model to', export_path

    builder = builder.SavedModelBuilder(export_path)
    builder.add_meta_graph_and_variables(sess, ["ID_TAG"],clear_devices=True)
    builder.save(as_text=True)
    print('Done exporting!')
    
    
### Part 3 Restore the graph using SavedModel API
tf.reset_default_graph()
dir(tf.contrib)
with tf.Session() as sess:  
    tf.saved_model.loader.load(sess, ["ID_TAG"], '_versions/%s'%version)
    graph = tf.get_default_graph()
        
    input_t = graph.get_tensor_by_name("input_example:0")
    output = graph.get_tensor_by_name("Sigmoid:0")
    
    print sess.run(output, feed_dict={input_t:np.random.random((10,5))})
    
    #THROWS THis ERROR
    # NotFoundError (see above for traceback): Function tf_map_func_7eefab31 is not defined.
    #      [[Node: ParallelMapDataset = ParallelMapDataset[Targuments=[], f=tf_map_func_7eefab31[],
    #      output_shapes=[[5]], output_types=[DT_FLOAT],
    #      _device="/job:localhost/replica:0/task:0/device:CPU:0"](TensorSliceDataset, num_parallel_calls)]]