75

I am working through the TensorFlow tutorial, which uses a "weird" format to upload the data. I would like to use the NumPy or pandas format for the data, so that I can compare it with scikit-learn results.

I get the digit recognition data from Kaggle: https://www.kaggle.com/c/digit-recognizer/data.

Here the code from the TensorFlow tutorial (which works fine):

# Stuff from tensorflow tutorial 
import tensorflow as tf

sess = tf.InteractiveSession()

x = tf.placeholder("float", shape=[None, 784])
y_ = tf.placeholder("float", shape=[None, 10])

W = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))

y = tf.nn.softmax(tf.matmul(x, W) + b)

cross_entropy = -tf.reduce_sum(y_ * tf.log(y))

train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)

Here I read the data, strip out the target variables and split the data into testing and training datasets (this all works fine):

# Read dataframe from training data
csvfile='train.csv'
from pandas import DataFrame, read_csv
df = read_csv(csvfile)

# Strip off the target data and make it a separate dataframe.
Target = df.label
del df["label"]

# Split data into training and testing sets
msk = np.random.rand(len(df)) < 0.8
dfTest = df[~msk]
TargetTest = Target[~msk]
df = df[msk]
Target = Target[msk]

# One hot encode the target
OHTarget=pd.get_dummies(Target)
OHTargetTest=pd.get_dummies(TargetTest)

Now, when I try to run the training step, I get a FailedPreconditionError:

for i in range(100):
    batch = np.array(df[i*50:i*50+50].values)
    batch = np.multiply(batch, 1.0 / 255.0)
    Target_batch = np.array(OHTarget[i*50:i*50+50].values)
    Target_batch = np.multiply(Target_batch, 1.0 / 255.0)
    train_step.run(feed_dict={x: batch, y_: Target_batch})

Here's the full error:

---------------------------------------------------------------------------
FailedPreconditionError                   Traceback (most recent call last)
<ipython-input-82-967faab7d494> in <module>()
      4     Target_batch = np.array(OHTarget[i*50:i*50+50].values)
      5     Target_batch = np.multiply(Target_batch, 1.0 / 255.0)
----> 6     train_step.run(feed_dict={x: batch, y_: Target_batch})

/Users/user32/anaconda/lib/python2.7/site-packages/tensorflow/python/framework/ops.pyc in run(self, feed_dict, session)
   1265         none, the default session will be used.
   1266     """
-> 1267     _run_using_default_session(self, feed_dict, self.graph, session)
   1268
   1269

/Users/user32/anaconda/lib/python2.7/site-packages/tensorflow/python/framework/ops.pyc in _run_using_default_session(operation, feed_dict, graph, session)
   2761                        "the operation's graph is different from the session's "
   2762                        "graph.")
-> 2763   session.run(operation, feed_dict)
   2764
   2765

/Users/user32/anaconda/lib/python2.7/site-packages/tensorflow/python/client/session.pyc in run(self, fetches, feed_dict)
    343
    344     # Run request and get response.
--> 345     results = self._do_run(target_list, unique_fetch_targets, feed_dict_string)
    346
    347     # User may have fetched the same tensor multiple times, but we

/Users/user32/anaconda/lib/python2.7/site-packages/tensorflow/python/client/session.pyc in _do_run(self, target_list, fetch_list, feed_dict)
    417         # pylint: disable=protected-access
    418         raise errors._make_specific_exception(node_def, op, e.error_message,
--> 419                                               e.code)
    420         # pylint: enable=protected-access
    421       raise e_type, e_value, e_traceback

FailedPreconditionError: Attempting to use uninitialized value Variable_1
     [[Node: gradients/add_grad/Shape_1 = Shape[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"](Variable_1)]]
Caused by op u'gradients/add_grad/Shape_1', defined at:
  File "/Users/user32/anaconda/lib/python2.7/runpy.py", line 162, in _run_module_as_main
    ...........

...which was originally created as op u'add', defined at:
  File "/Users/user32/anaconda/lib/python2.7/runpy.py", line 162, in _run_module_as_main
    "__main__", fname, loader, pkg_name)
[elided 17 identical lines from previous traceback]
  File "/Users/user32/anaconda/lib/python2.7/site-packages/IPython/core/interactiveshell.py", line 3066, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-45-59183d86e462>", line 1, in <module>
    y = tf.nn.softmax(tf.matmul(x,W) + b)
  File "/Users/user32/anaconda/lib/python2.7/site-packages/tensorflow/python/ops/math_ops.py", line 403, in binary_op_wrapper
    return func(x, y, name=name)
  File "/Users/user32/anaconda/lib/python2.7/site-packages/tensorflow/python/ops/gen_math_ops.py", line 44, in add
    return _op_def_lib.apply_op("Add", x=x, y=y, name=name)
  File "/Users/user32/anaconda/lib/python2.7/site-packages/tensorflow/python/ops/op_def_library.py", line 633, in apply_op
    op_def=op_def)
  File "/Users/user32/anaconda/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1710, in create_op
    original_op=self._default_original_op, op_def=op_def)
  File "/Users/user32/anaconda/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 988, in __init__
    self._traceback = _extract_stack()

Any ideas as to how I can fix this?

nbro
  • 15,395
  • 32
  • 113
  • 196
user3654387
  • 2,240
  • 4
  • 19
  • 20

11 Answers11

94

The FailedPreconditionError arises because the program is attempting to read a variable (named "Variable_1") before it has been initialized. In TensorFlow, all variables must be explicitly initialized, by running their "initializer" operations. For convenience, you can run all of the variable initializers in the current session by executing the following statement before your training loop:

tf.initialize_all_variables().run()

Note that this answer assumes that, as in the question, you are using tf.InteractiveSession, which allows you to run operations without specifying a session. For non-interactive uses, it is more common to use tf.Session, and initialize as follows:

init_op = tf.initialize_all_variables()

sess = tf.Session()
sess.run(init_op)
mrry
  • 125,488
  • 26
  • 399
  • 400
  • 16
    Use `tf.global_variables_initializer` instead because `tf.initialize_all_variables()` will be deprecated. – Beginner Jan 31 '18 at 16:14
64

tf.initialize_all_variables() is deprecated. Instead initialize tensorflow variables with: 

tf.global_variables_initializer()

A common example usage is:

with tf.Session() as sess:
     sess.run(tf.global_variables_initializer())
nbro
  • 15,395
  • 32
  • 113
  • 196
user3144836
  • 4,060
  • 3
  • 27
  • 27
  • what's the difference between sess.run(tf.global_variables_initializer()) and tf.global_variables_initializer()? you often see the latter – dv3 Feb 14 '17 at 09:42
  • 2
    @dv3 like everything else in TensorFlow, there is a difference between creating the operation and running it. I think that's why "initialize_all_variables" was replaced with "global_variables_initializer": The name "initialize_all_variables" sounds like it does it, when it only creates the operation. – mdaoust Apr 17 '17 at 16:25
  • pls note that the method is now tf.initialize_all_variables() – jesses.co.tt Apr 27 '17 at 15:12
  • 1
    @jesses.co.tt - answer is correct; `tf.initialize_all_variables()` is the one deprecated. See: https://www.tensorflow.org/api_docs/python/tf/initialize_all_variables – ruhong Oct 03 '18 at 00:15
  • @user3144836's answer worked perfectly. In my case, I encapsulated the entire code block under the with routine, and that purged all the errors. Thanks. – God Bennett Aug 05 '19 at 00:25
16

From official documentation, FailedPreconditionError

This exception is most commonly raised when running an operation that reads a tf.Variable before it has been initialized.

In your case the error even explains what variable was not initialized: Attempting to use uninitialized value Variable_1. One of the TF tutorials explains a lot about variables, their creation/initialization/saving/loading

Basically to initialize the variable you have 3 options:

I almost always use the first approach. Remember you should put it inside a session run. So you will get something like this:

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

If your are curious about more information about variables, read this documentation to know how to report_uninitialized_variables and check is_variable_initialized.

Salvador Dali
  • 214,103
  • 147
  • 703
  • 753
4

You have to initialize variables before using them.​

If you try to evaluate the variables before initializing them you'll run into: FailedPreconditionError: Attempting to use uninitialized value tensor.

The easiest way is initializing all variables at once using: tf.global_variables_initializer()

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

You use sess.run(init) to run the initializer, without fetching any value.

To initialize only a subset of variables, you use tf.variables_initializer() listing the variables:

var_ab = tf.variables_initializer([a, b], name="a_and_b")
with tf.Session() as sess:
    sess.run(var_ab)

You can also initialize each variable separately using tf.Variable.initializer

# create variable W as 784 x 10 tensor, filled with zeros
W = tf.Variable(tf.zeros([784,10])) with tf.Session() as sess:
    sess.run(W.initializer)
Tapio
  • 1,502
  • 1
  • 12
  • 24
prosti
  • 42,291
  • 14
  • 186
  • 151
3

I got this error message from a completely different case. It seemed that the exception handler in tensorflow raised it. You can check each row in the Traceback. In my case, it happened in tensorflow/python/lib/io/file_io.py, because this file contained a different bug, where self.__mode and self.__name weren't initialized, and it needed to call self._FileIO__mode, and self_FileIO__name instead.

nbro
  • 15,395
  • 32
  • 113
  • 196
Tal
  • 127
  • 1
  • 7
3

Different use case, but set your session as the default session did the trick for me:

with sess.as_default():
    result = compute_fn([seed_input,1])

This is one of these mistakes that is so obvious, once you have solved it.

My use-case is the following:
1) store keras VGG16 as tensorflow graph
2) load kers VGG16 as a graph
3) run tf function on the graph and get: 

FailedPreconditionError: Attempting to use uninitialized value block1_conv2/bias
     [[Node: block1_conv2/bias/read = Identity[T=DT_FLOAT, _class=["loc:@block1_conv2/bias"], _device="/job:localhost/replica:0/task:0/device:GPU:0"](block1_conv2/bias)]]
     [[Node: predictions/Softmax/_7 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_168_predictions/Softmax", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]
layser
  • 2,038
  • 3
  • 15
  • 14
3

The FailedPreconditionError comes because the session is trying to read a variable that hasn"t been initialized.

As of Tensorflow version 1.11.0, you need to take this :

init_op = tf.global_variables_initializer()

sess = tf.Session()
sess.run(init_op)
LaSul
  • 2,231
  • 1
  • 20
  • 36
3

Tensorflow 2.0 Compatible Answer: In Tensorflow Version >= 2.0, the command for Initializing all the Variables if we use Graph Mode, to fix the FailedPreconditionError is shown below:

tf.compat.v1.global_variables_initializer

This is just a shortcut for variables_initializer(global_variables())

It returns an Op that initializes global variables in the graph.

  • I still see this error even after initializing in 2.3 when using with `tf.data`. https://stackoverflow.com/questions/64094061/failedpreconditionerror-getnext-failed-because-the-iterator-has-not-been-init – Nitin Sep 28 '20 at 06:40
3

With tensorflow version >3.2 you may use this command:

x1 = tf.Variable(5)
y1 = tf.Variable(3)

z1 = x1 + y1

init = tf.compat.v1.global_variables_initializer()
with tf.compat.v1.Session() as sess:
    init.run()
    print(sess.run(z1))

The output: 8 will be displayed.

Piyush Katariya
  • 31
  • 4
2

When I had this issue with tf.train.string_input_producer() and tf.train.batch() initializing the local variables before I started the Coordinator solved the problem. I had been getting the error when I initialized the local variables after starting the Coordinator.

nbro
  • 15,395
  • 32
  • 113
  • 196
Eric Antoine Scuccimarra
  • 375
  • 2
  • 7
2

Possibly something has changed in recent TensorFlow builds, because for me, running

sess = tf.Session()
sess.run(tf.local_variables_initializer())

before fitting any models seems to do the trick. Most older examples and comments seem to suggest tf.global_variables_initializer().

wordsforthewise
  • 13,746
  • 5
  • 87
  • 117