FailedPreconditionError: lack of intialization of Keras variables when using with TensorFlow

[dumkar]

I am new to Keras and just installed it (with pip3) to use with TensorFlow (1.0.0). I am trying to follow the Keras+TensorFlow tutorial.

When running the code, it stops at

train_step.run(feed_dict={img: batch[0], labels: batch[1]})

and throws the error below. I figured out it is because variables are not initialized and fixed it by inserting (see #4623):

keras.backend.get_session().run(tf.global_variables_initializer())

I decided to post it here since I was wondering if this is a general issue with Keras (as this is a rather simple example) regarding the update to TensorFlow 1.0.0 or something specific to my setup?

The error:

---

FailedPreconditionError Traceback (most recent call last) C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in _do_call(self, fn, args) 1021 try: -> 1022 return fn(args) 1023 except errors.OpError as e:

C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in _run_fn(session, feed_dict, fetch_list, target_list, options, run_metadata) 1003 feed_dict, fetch_list, target_list, -> 1004 status, run_metadata) 1005

C:\Users\dumon\Anaconda3\lib\contextlib.py in exit(self, type, value, traceback) 65 try: ---> 66 next(self.gen) 67 except StopIteration:

C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\framework\errors_impl.py in raise_exception_on_not_ok_status() 468 compat.as_text(pywrap_tensorflow.TF_Message(status)), --> 469 pywrap_tensorflow.TF_GetCode(status)) 470 finally:

FailedPreconditionError: Attempting to use uninitialized value dense_1_W [[Node: dense_1_W/read = IdentityT=DT_FLOAT, _class=["loc:@dense_1_W"], _device="/job:localhost/replica:0/task:0/cpu:0"]]

During handling of the above exception, another exception occurred:

FailedPreconditionError Traceback (most recent call last)

in () 9 batch= mnist_data.train.next_batch(50) 10 train_step.run(feed_dict={img: batch[0], ---> 11 labels: batch[1]}) C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\framework\ops.py in run(self, feed_dict, session) 1586 none, the default session will be used. 1587 """ -> 1588 _run_using_default_session(self, feed_dict, self.graph, session) 1589 1590 C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\framework\ops.py in _run_using_default_session(operation, feed_dict, graph, session) 3830 "the operation's graph is different from the session's " 3831 "graph.") -> 3832 session.run(operation, feed_dict) 3833 3834 C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in run(self, fetches, feed_dict, options, run_metadata) 765 try: 766 result = self._run(None, fetches, feed_dict, options_ptr, --> 767 run_metadata_ptr) 768 if run_metadata: 769 proto_data = tf_session.TF_GetBuffer(run_metadata_ptr) C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in _run(self, handle, fetches, feed_dict, options, run_metadata) 963 if final_fetches or final_targets: 964 results = self._do_run(handle, final_targets, final_fetches, --> 965 feed_dict_string, options, run_metadata) 966 else: 967 results = [] C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata) 1013 if handle is None: 1014 return self._do_call(_run_fn, self._session, feed_dict, fetch_list, -> 1015 target_list, options, run_metadata) 1016 else: 1017 return self._do_call(_prun_fn, self._session, handle, feed_dict, C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in _do_call(self, fn, *args) 1033 except KeyError: 1034 pass -> 1035 raise type(e)(node_def, op, message) 1036 1037 def _extend_graph(self): FailedPreconditionError: Attempting to use uninitialized value dense_1_W [[Node: dense_1_W/read = Identity[T=DT_FLOAT, _class=["loc:@dense_1_W"], _device="/job:localhost/replica:0/task:0/cpu:0"](dense_1_W)]] Caused by op 'dense_1_W/read', defined at: File "C:\Users\dumon\Anaconda3\lib\runpy.py", line 184, in _run_module_as_main "__main__", mod_spec) File "C:\Users\dumon\Anaconda3\lib\runpy.py", line 85, in _run_code exec(code, run_globals) File "C:\Users\dumon\Anaconda3\lib\site-packages\ipykernel\__main__.py", line 3, in app.launch_new_instance() File "C:\Users\dumon\Anaconda3\lib\site-packages\traitlets\config\application.py", line 653, in launch_instance app.start() File "C:\Users\dumon\Anaconda3\lib\site-packages\ipykernel\kernelapp.py", line 474, in start ioloop.IOLoop.instance().start() File "C:\Users\dumon\Anaconda3\lib\site-packages\zmq\eventloop\ioloop.py", line 162, in start super(ZMQIOLoop, self).start() File "C:\Users\dumon\Anaconda3\lib\site-packages\tornado\ioloop.py", line 887, in start handler_func(fd_obj, events) File "C:\Users\dumon\Anaconda3\lib\site-packages\tornado\stack_context.py", line 275, in null_wrapper return fn(*args, **kwargs) File "C:\Users\dumon\Anaconda3\lib\site-packages\zmq\eventloop\zmqstream.py", line 440, in _handle_events self._handle_recv() File "C:\Users\dumon\Anaconda3\lib\site-packages\zmq\eventloop\zmqstream.py", line 472, in _handle_recv self._run_callback(callback, msg) File "C:\Users\dumon\Anaconda3\lib\site-packages\zmq\eventloop\zmqstream.py", line 414, in _run_callback callback(*args, **kwargs) File "C:\Users\dumon\Anaconda3\lib\site-packages\tornado\stack_context.py", line 275, in null_wrapper return fn(*args, **kwargs) File "C:\Users\dumon\Anaconda3\lib\site-packages\ipykernel\kernelbase.py", line 276, in dispatcher return self.dispatch_shell(stream, msg) File "C:\Users\dumon\Anaconda3\lib\site-packages\ipykernel\kernelbase.py", line 228, in dispatch_shell handler(stream, idents, msg) File "C:\Users\dumon\Anaconda3\lib\site-packages\ipykernel\kernelbase.py", line 390, in execute_request user_expressions, allow_stdin) File "C:\Users\dumon\Anaconda3\lib\site-packages\ipykernel\ipkernel.py", line 196, in do_execute res = shell.run_cell(code, store_history=store_history, silent=silent) File "C:\Users\dumon\Anaconda3\lib\site-packages\ipykernel\zmqshell.py", line 501, in run_cell return super(ZMQInteractiveShell, self).run_cell(*args, **kwargs) File "C:\Users\dumon\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 2717, in run_cell interactivity=interactivity, compiler=compiler, result=result) File "C:\Users\dumon\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 2821, in run_ast_nodes if self.run_code(code, result): File "C:\Users\dumon\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 2881, in run_code exec(code_obj, self.user_global_ns, self.user_ns) File "", line 4, in x = Dense(128, activation='relu')(img) # fully-connected layer with 128 units and ReLU activation File "C:\Users\dumon\Anaconda3\lib\site-packages\keras\engine\topology.py", line 546, in __call__ self.build(input_shapes[0]) File "C:\Users\dumon\Anaconda3\lib\site-packages\keras\layers\core.py", line 798, in build constraint=self.W_constraint) File "C:\Users\dumon\Anaconda3\lib\site-packages\keras\engine\topology.py", line 418, in add_weight weight = initializer(shape, name=name) File "C:\Users\dumon\Anaconda3\lib\site-packages\keras\initializations.py", line 66, in glorot_uniform return uniform(shape, s, name=name) File "C:\Users\dumon\Anaconda3\lib\site-packages\keras\initializations.py", line 33, in uniform return K.random_uniform_variable(shape, -scale, scale, name=name) File "C:\Users\dumon\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py", line 635, in random_uniform_variable return variable(value, dtype=dtype, name=name) File "C:\Users\dumon\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py", line 259, in variable v = tf.Variable(value, dtype=_convert_string_dtype(dtype), name=name) File "C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\ops\variables.py", line 226, in __init__ expected_shape=expected_shape) File "C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\ops\variables.py", line 344, in _init_from_args self._snapshot = array_ops.identity(self._variable, name="read") File "C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\ops\gen_array_ops.py", line 1490, in identity result = _op_def_lib.apply_op("Identity", input=input, name=name) File "C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 763, in apply_op op_def=op_def) File "C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\framework\ops.py", line 2395, in create_op original_op=self._default_original_op, op_def=op_def) File "C:\Users\dumon\Anaconda3\lib\site-packages\tensorflow\python\framework\ops.py", line 1264, in __init__ self._traceback = _extract_stack() FailedPreconditionError (see above for traceback): Attempting to use uninitialized value dense_1_W [[Node: dense_1_W/read = Identity[T=DT_FLOAT, _class=["loc:@dense_1_W"], _device="/job:localhost/replica:0/task:0/cpu:0"](dense_1_W)]]

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs, but feel free to re-open it if needed.

[DavidWilt]

I am as of July 11, 2017, experiencing this bug. It first occurred when I was attempting to train a model, but it now arises when the workaround code above is executed: keras.backend.get_session().run(tf.global_variables_initializer())

Error message below. What else do you need?

Thanks.

ensorflow.python.framework.errors_impl.FailedPreconditionError: Attempting to use uninitialized value dense_1/Variable [[Node: dense_1/Variable/read = IdentityT=DT_FLOAT, _class=["loc:@dense_1/Variable"], _device="/job:localhost/replica:0/task:0/cpu:0"]] Caused by op u'dense_1/Variable/read', defined at: File "/Applications/WingIDE.app/Contents/Resources/bin/wingdb.py", line 978, in main() File "/Applications/WingIDE.app/Contents/Resources/bin/wingdb.py", line 918, in main netserver.abstract.kFileSystemEncoding, orig_sys_path) File "/Applications/WingIDE.app/Contents/Resources/bin/wingdb.py", line 766, in DebugFile exit_code = server.Run(filename, sys.argv)

[ylmeng]

I have the same problem. I have a custom layer which works fine in some models, but fails with this message (similar to above) in other models. Totally annoying.

[why2face]

Got the same error while trying to use tensorflow-gpu as backend in keras. Though it worked well using cpu before. How could this make a difference?

[AniketDhar]

Facing the same issue while training Keras model with custom kernel initializers. Also happens if I add BatchNormalization in the model. I already tried tf.global_variables_initializer() before fit, but that did not help. Any suggestions or workarounds?

[drsxr]

Also having the same problem - happens in the BatchNorm layer. Took one version of code, ran it on GPU:0, no problem. Copied the code, Ran it on GPU:1, changed a few of the hyperparameters (learning rate, # of epochs) and get a FailedPreconditionError. Very inconstant, but once it happens in one of my Jupyter Notebooks, it seems reproducible there. Using Keras 2.1.3 and TF 1.8

[alexattia]

I have the same issue, any suggestions?

[evertonaleixo]

I initialize the variables with the following code, and its work for me:

K.set_session(tf.Session(graph=model.output.graph)) init = K.tf.global_variables_initializer() K.get_session().run(init)

where K is from 'from keras import backend as K'. tf is from 'import tensorflow as tf'. And 'model' is my keras model. I add this code after compile the model.

[YTSKanuri]

The only solution that worked for me when using notebook is:-

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    hist = model.fit_generator(
    train_datagen, steps_per_epoch=STEPS, epochs=EPOCHS, verbose=1,
    validation_data=(x_valid, y_valid),
    callbacks = callbacks_list)

[nateGeorge]

For me, I had to use local_variables_initilaizer() -- global_variables_initializer() wouldn't work.

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

[novioleo]

same error in latest keras.

[novioleo]

I fixed it with set different graph in session. if there are multi-models in the same project,use the tensorflow default graph to init a new session,and a definitely new graph for tensorflow model.

[edufonseca]

@novioleo could you please share the code snippet that fixed this? thanks

[novioleo]

@novioleo could you please share the code snippet that fixed this? thanks

i'm not exactly sure~but you can have a reference:

# model from **keras** please use the default graph **always**.
# model from tensorflow need to use a totally new graph
default_graph = tf.get_default_graph()
with default_graph.as_default():
    self.sess_1 = tf.Session(config=self.config)
     K.set_session(self.sess_1)
     with self.sess.as_default():
         self.model = modellib.MaskRCNN(mode="inference", model_dir=self.log_dir, config=InferenceConfig())
         self.model.load_weights(self.model_file, by_name=True)

with graph.as_default():
    self.x_ = tf.placeholder(tf.float32, [None, self.img_size])
    self.x_image = tf.reshape(self.x_, [-1, self.img_height, self.img_width, 3])
    self.enhanced = resnet(self.x_image)
    self.sess_2 = tf.Session(config=self.config)
    with self.sess_2.as_default():
        saver = tf.train.Saver()
        saver.restore(self.sess_2, "./path/to/your/model")

i exacted my this from my code,there could be some errors,please fix it by yourself~ when you need the use the model to predict,justwith self.which_session_you_want_to_use:. i suggest you guys can make a model into a class for a better management.

[menon92]

This solve my issue

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    # do others task

[LeeKeyu]

I fixed it with set different graph in session. if there are multi-models in the same project,use the tensorflow default graph to init a new session,and a definitely new graph for tensorflow model.

My issue is also caused by using multi-models in one project (tensorflow model + Keras model). Thank @novioleo for the answer. My issue was solved by initializing the Keras model using a new session defined in the default graph

default_graph = tf.get_default_graph()
with default_graph.as_default():
    self.sess_keras = tf.Session()
    global model
    model = Model()     # keras model

and use this new session during prediction with the keras model:

with self.sess_keras.as_default():
    test_logits = model.predict()