Tensorflow reading data from tfrecords into mini batches properly

Question

I am trying to convert data from csv to tfrecords then read it out in mini batches and do a simple MLP but I am getting some errors that I can't figure out.

RuntimeError: Attempted to use a closed Session.

Followed by:

TypeError: The value of a feed cannot be a tf.Tensor object. Acceptable feed values include Python scalars, strings, lists, or numpy ndarrays.

I'm guessing the shuffle batch queue is closing and no longer feeding the expected data. Also, I think I am missing a step going from the shuffle queue to the feed dict. Any ideas how to make this work or a better way of doing it?

Here is my code:

import numpy as np
import tensorflow as tf
import pandas

filename = "test.tfrecords"
writer = tf.python_io.TFRecordWriter(filename)

csv = pandas.read_csv("TRAINING.csv").values
with tf.python_io.TFRecordWriter(filename) as writer:
    for row in csv:
        features, label = row[:-1], row[-1]
        example = tf.train.Example()
        example.features.feature["avgs"].float_list.value.extend(features)
        example.features.feature["pdiff"].float_list.value.append(label)
        writer.write(example.SerializeToString())

def read_and_decode_single_example(filename):
    filename_queue = tf.train.string_input_producer([filename],
                                                   num_epochs=None)
    reader = tf.TFRecordReader()
    _, serialized_example = reader.read(filename_queue)

    features = tf.parse_single_example(
        serialized_example,
        features={
            'pdiff': tf.FixedLenFeature([], np.float32),
            'avgs': tf.FixedLenFeature([14], np.float32)})

    pdiff = features['pdiff']
    avgs = features['avgs']

    return avgs, pdiff

avgs, pdiff = read_and_decode_single_example(filename)

avgs_batch, pdiff_batch = tf.train.shuffle_batch(
    [avgs, pdiff], batch_size=200,
    capacity=2000,
    min_after_dequeue=600)


n_features = 14
batch_size = 50
hidden_units = 7
lr = .03

X = tf.placeholder(tf.float32,[None,n_features])
Y = tf.placeholder(tf.float32,[None])

W = tf.Variable(tf.truncated_normal([n_features,hidden_units]))
Wout = tf.Variable(tf.truncated_normal([hidden_units,1]))

b = tf.Variable(tf.zeros([hidden_units]))
bout = tf.Variable(tf.zeros([1]))


hidden1 = tf.nn.sigmoid(tf.matmul(X,W) + b)
pred = tf.matmul(hidden1,Wout) + bout

loss = tf.reduce_sum(tf.pow(pred - Y,2))
optimizer = tf.train.AdamOptimizer(lr).minimize(loss)


 with tf.Session() as sess:
    init = tf.global_variables_initializer()
    sess.run(init)
    tf.train.start_queue_runners(sess=sess)

    for step in range(1000):
        _, loss_val = sess.run([optimizer,loss],
                feed_dict={X: avgs_batch, Y: pdiff_batch} )

Stack Trace:

ERROR:tensorflow:Exception in QueueRunner: Attempted to use a closed Session.
Traceback (most recent call last):
ERROR:tensorflow:Exception in QueueRunner: Attempted to use a closed Session.
  File "tf_tb.py", line 110, in <module>
    feed_dict={X: avgs_batch, Y: pdiff_batch} )
  File "/Users/mnaymik/.envs/venv/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 766, in run
    run_metadata_ptr)
  File "/Users/mnaymik/.envs/venv/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 924, in _run
Exception in thread Thread-1:
Traceback (most recent call last):
  File "/usr/local/Cellar/python3/3.5.2_3/Frameworks/Python.framework/Versions/3.5/lib/python3.5/threading.py", line 914, in _bootstrap_inner
    self.run()
  File "/usr/local/Cellar/python3/3.5.2_3/Frameworks/Python.framework/Versions/3.5/lib/python3.5/threading.py", line 862, in run
    self._target(*self._args, **self._kwargs)
  File "/Users/mnaymik/.envs/venv/lib/python3.5/site-packages/tensorflow/python/training/queue_runner_impl.py", line 234, in _run
    sess.run(enqueue_op)
  File "/Users/mnaymik/.envs/venv/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 766, in run
    run_metadata_ptr)
  File "/Users/mnaymik/.envs/venv/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 902, in _run
    raise RuntimeError('Attempted to use a closed Session.')
RuntimeError: Attempted to use a closed Session.

Exception in thread Thread-2:
Traceback (most recent call last):
  File "/usr/local/Cellar/python3/3.5.2_3/Frameworks/Python.framework/Versions/3.5/lib/python3.5/threading.py", line 914, in _bootstrap_inner
    self.run()
  File "/usr/local/Cellar/python3/3.5.2_3/Frameworks/Python.framework/Versions/3.5/lib/python3.5/threading.py", line 862, in run
    self._target(*self._args, **self._kwargs)
  File "/Users/mnaymik/.envs/venv/lib/python3.5/site-packages/tensorflow/python/training/queue_runner_impl.py", line 234, in _run
    sess.run(enqueue_op)
  File "/Users/mnaymik/.envs/venv/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 766, in run
    run_metadata_ptr)
  File "/Users/mnaymik/.envs/venv/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 902, in _run
    raise RuntimeError('Attempted to use a closed Session.')
RuntimeError: Attempted to use a closed Session.

    raise TypeError('The value of a feed cannot be a tf.Tensor object. '
TypeError: The value of a feed cannot be a tf.Tensor object. Acceptable feed values include Python scalars, strings, lists, or numpy ndarrays.

Answer 1

Placeholders are one way of getting data into your model. Queue's are another. You can override the value in a tensor generated by a queue runner like any other tensor (e.g. placeholders), but you can't feed the results from a tensor into a placeholder in the same graph/session run.

In other words, rather than creating placeholders, just use the output of your tf.train.shuffle_batch call:

X = avgs_batch
Y = pdiff_batch

(or replace all references to X and Y with avgs_batch and pdiff_batch respectively)