Loading data in TensorFlow, using low-level queuing mechanism

feeder.py

import tensorflow as tf

class Feeder:
	def __init__(self):
		self.sess = tf.Session() # session to run operations
		self.coordinator = tf.train.Coordinator() # create coordinator for threads

		self.placeholder = tf.placeholder(tf.int32, shape=(None, None), name='sentence')
		self.queue = tf.FIFOQueue(5, tf.int32, name='input_queue') # hold 5 elements
		self.enqueue_op = queue.enqueue(placeholder) # push op
		self.sentence = queue.dequeue()  # pop op

	# The main procedure
	def enqueue(batch_size=5, num_batches=5):
		while not self.coordinator.should_stop():
			# 1 and 2: read data from dist and preprocess
			examples = [self.get_next_example() for _ in range(batch_size * num_batches)]
			# 3: Make batches
			batches = [[examples[i: i + batch_size]] for i in range(0, len(examples), batch_size)]
			# 4: Insert the batches in the queue
			[sess.run(enqueue_op, feed_dict=dict(zip(placeholder, batch))) for batch in batches]

	# Repeat the same process over and over
	def start_threads(self):
		thread = threading.Thread(name='background', target=self.enqueue)
        	thread.daemon = True
        	thread.start()