You must feed a value for placeholder tensor 'c_1' with dtype float and shape [1]

Question

I read some issue about same problem, but it looks like my issue some different. I want to freeze graph and then use it.

Here simple example how I do this. First, I create session and save both checkpoint and GraphDef:

a = tf.Variable(tf.constant(1.), name='a')
b = tf.Variable(tf.constant(2.), name='b')
c = tf.placeholder(tf.float32, shape =[1], name="c")
add = tf.add(a, b, 'sum')
add2 = tf.add(add, c, 'sum2')

dir_path = "<full_path>/simple_store"

with tf.Session() as sess:
    tf.initialize_all_variables().run()
    sess.run([add2], feed_dict={c:[7.]})
    tf.train.Saver().save(sess, dir_path + "/" + 'simple.ckpt')
    tf.train.write_graph(graph_def=sess.graph.as_graph_def(), logdir=dir_path, name='simple_as_text.pb')

Then I use bazel tool for freezing such way:

../tensorflow/bazel-bin/tensorflow/python/tools/freeze_graph --input_graph=simple_store/simple_as_text.pb --input_checkpoint=simple_store/simple.ckpt --output_graph=simple_store/freeze_out.pb --output_node_names=sum2

Then I load freeze_out.pb in Python and try run:

import tensorflow as tf
from tensorflow.core.framework import graph_pb2, cost_graph_pb2

graph_def = graph_pb2.GraphDef()
d = None

c = tf.placeholder(tf.float32, shape=[1], name="c")
feed_dict = {c: [5.]}

with tf.Session() as session:
    print("load graph")
    with open("<somepath>/simple_store/freeze_out.pb", "rb") as f:
        graph_def.ParseFromString(f.read())
        d = tf.import_graph_def(graph_def, return_elements=["sum2:0"], name='')
    print(session.run([d[0]], feed_dict=feed_dict))

And finally I get following error:

---------------------------------------------------------------------------
InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-3-6345b17fba3b> in <module>()
      8         d = tf.import_graph_def(graph_def, return_elements=["sum2:0"], name='')
      9     tf.initialize_all_variables().run()
---> 10     print(session.run([d[0]], feed_dict=feed_dict))

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in run(self, fetches, feed_dict, options, run_metadata)
    370     try:
    371       result = self._run(None, fetches, feed_dict, options_ptr,
--> 372                          run_metadata_ptr)
    373       if run_metadata:
    374         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in _run(self, handle, fetches, feed_dict, options, run_metadata)
    634     try:
    635       results = self._do_run(handle, target_list, unique_fetches,
--> 636                              feed_dict_string, options, run_metadata)
    637     finally:
    638       # The movers are no longer used. Delete them.

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)
    706     if handle is None:
    707       return self._do_call(_run_fn, self._session, feed_dict, fetch_list,
--> 708                            target_list, options, run_metadata)
    709     else:
    710       return self._do_call(_prun_fn, self._session, handle, feed_dict,

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in _do_call(self, fn, *args)
    726         except KeyError:
    727           pass
--> 728       raise type(e)(node_def, op, message)
    729 
    730   def _extend_graph(self):

InvalidArgumentError: You must feed a value for placeholder tensor 'c_1' with dtype float and shape [1]
     [[Node: c_1 = Placeholder[dtype=DT_FLOAT, shape=[1], _device="/job:localhost/replica:0/task:0/cpu:0"]()]]
Caused by op u'c_1', defined at:
  File "/usr/lib/python2.7/runpy.py", line 174, in _run_module_as_main
    "__main__", fname, loader, pkg_name)
  File "/usr/lib/python2.7/runpy.py", line 72, in _run_code
    exec code in run_globals
  File "/home/artem/.local/lib/python2.7/site-packages/ipykernel/__main__.py", line 3, in <module>
    app.launch_new_instance()
  File "/home/artem/.local/lib/python2.7/site-packages/traitlets/config/application.py", line 596, in launch_instance
    app.start()
  File "/home/artem/.local/lib/python2.7/site-packages/ipykernel/kernelapp.py", line 442, in start
    ioloop.IOLoop.instance().start()
  File "/home/artem/.local/lib/python2.7/site-packages/zmq/eventloop/ioloop.py", line 162, in start
    super(ZMQIOLoop, self).start()
  File "/home/artem/.local/lib/python2.7/site-packages/tornado/ioloop.py", line 887, in start
    handler_func(fd_obj, events)
  File "/home/artem/.local/lib/python2.7/site-packages/tornado/stack_context.py", line 275, in null_wrapper
    return fn(*args, **kwargs)
  File "/home/artem/.local/lib/python2.7/site-packages/zmq/eventloop/zmqstream.py", line 440, in _handle_events
    self._handle_recv()
  File "/home/artem/.local/lib/python2.7/site-packages/zmq/eventloop/zmqstream.py", line 472, in _handle_recv
    self._run_callback(callback, msg)
  File "/home/artem/.local/lib/python2.7/site-packages/zmq/eventloop/zmqstream.py", line 414, in _run_callback
    callback(*args, **kwargs)
  File "/home/artem/.local/lib/python2.7/site-packages/tornado/stack_context.py", line 275, in null_wrapper
    return fn(*args, **kwargs)
  File "/home/artem/.local/lib/python2.7/site-packages/ipykernel/kernelbase.py", line 276, in dispatcher
    return self.dispatch_shell(stream, msg)
  File "/home/artem/.local/lib/python2.7/site-packages/ipykernel/kernelbase.py", line 228, in dispatch_shell
    handler(stream, idents, msg)
  File "/home/artem/.local/lib/python2.7/site-packages/ipykernel/kernelbase.py", line 391, in execute_request
    user_expressions, allow_stdin)
  File "/home/artem/.local/lib/python2.7/site-packages/ipykernel/ipkernel.py", line 199, in do_execute
    shell.run_cell(code, store_history=store_history, silent=silent)
  File "/home/artem/.local/lib/python2.7/site-packages/IPython/core/interactiveshell.py", line 2705, in run_cell
    interactivity=interactivity, compiler=compiler, result=result)
  File "/home/artem/.local/lib/python2.7/site-packages/IPython/core/interactiveshell.py", line 2809, in run_ast_nodes
    if self.run_code(code, result):
  File "/home/artem/.local/lib/python2.7/site-packages/IPython/core/interactiveshell.py", line 2869, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-3-6345b17fba3b>", line 8, in <module>
    d = tf.import_graph_def(graph_def, return_elements=["sum2:0"], name='')
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/importer.py", line 274, in import_graph_def
    op_def=op_def)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 2260, in create_op
    original_op=self._default_original_op, op_def=op_def)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 1230, in __init__
    self._traceback = _extract_stack()

What I did wrong? How I should correct this?

Answer 1

In my case when I tried the above method, the node automatically got renamed. So, here's what worked for me. First I printed out all the nodes using

[n.name for n in tf.get_default_graph().as_graph_def().node] 

I had not named my placeholders while training my model but looking at output of above code I could make out that they were automatically named 'Placeholder' and 'Placeholder_1'

Then I used following lines to get tensors

x = tf.get_default_graph().get_tensor_by_name("Placeholder:0")
y = tf.get_default_graph().get_tensor_by_name("Placeholder_1:0")

This gave me the required placeholders. In case for above question, after importing the graph def, just doing c = tf.get_default_graph().get_tensor_by_name("c:0") should work.

Answer 2

The tf.import_graph_def() function maintains the structure of the imported graph, unless you pass the input_map argument.

In the original graph you passed to freeze_graph, the tensor named "sum2:0" depends on a placeholder operation called "c" which is in the same graph. When you import the frozen graph, TensorFlow first imports the node named "c"

When you import the frozen graph using tf.import_graph_def(), TensorFlow first imports the node named "c" in freeze_out.pb. However, because you have created another node named "c" in your second program, the imported node is renamed to "c_1" (an automatically generated unique name) and the imported version of "sum_2" is rewritten to depend on "c_1". Notably, it does not depend on the placeholder that you are feeding (which is named "c").

There are two solutions. The more straightforward solution is to extract the previously created placeholder from the imported graph, rather than creating a new one. You can do this by adding "c:0" to the list of return_elements:

graph_def = tf.GraphDef()

with open("<somepath>/simple_store/freeze_out.pb", "rb") as f:
    graph_def.ParseFromString(f.read())

# Also extract the placeholder "c" from the imported graph.
c, d = tf.import_graph_def(graph_def, return_elements=["c:0", "sum2:0"])

with tf.Session() as session:
    print(session.run([d[0]], feed_dict={c: [5.]}))

Alternatively, you can remap the placeholder in the imported graph to use the placeholder in your new graph. (There is not much point in doing this substitution, but it can be useful when the new graph is more complex and includes some new preprocessing, for example.) This uses the input_map argument to tf.import_graph_def():

graph_def = tf.GraphDef()

# Create a new placeholder that we will map into the imported graph.
# (N.B. This has no advantage, but could be useful if `c` were a more interesting
# function.)
c = tf.placeholder(tf.float32, shape=[1], name="c")
feed_dict = {c: [5.]}

with open("<somepath>/simple_store/freeze_out.pb", "rb") as f:
    graph_def.ParseFromString(f.read())

# Also remap the placeholder in the imported graph to use the placeholder created
# above. Notice that the syntax is like the feed_dict, but this performs a static 
# remapping of one tensor to another at graph construction time.
d = tf.import_graph_def(graph_def, input_map={"c:0": c}, return_elements=["sum2:0"])

with tf.Session() as session:
    print(session.run([d[0]], feed_dict={c: [5.]}))