How to restore checkpoint in TensorFlow inside ipython or anaconda

Question

I use tensorflow 0.9.
I want to save my model and then restore it.
I simply add tf.train.Saver() to save and restore my training variables.

This is my code:

import tensorflow as tf
import input_data
import os

checkpoint_dir='./ckpt_dir/'

mnist = input_data.read_data_sets("MNIST_data", one_hot = True)

x = tf.placeholder(tf.float32, shape = [None , 784])
y_ = tf.placeholder(tf.float32, [None, 10])

sess = tf.InteractiveSession()

def load_model(sess, saver, checkpoint_dir ):

ckpt = tf.train.get_checkpoint_state(checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
print(ckpt.model_checkpoint_path)

saver.restore(sess, ckpt.model_checkpoint_path)

else:
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
sess.run(init)
return

def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev = 0.1)
return tf.Variable(initial)

def bias_variable(shape):
initial = tf.constant(0.1, shape= shape)
return tf.Variable(initial)

def conv2d(x, W):
return tf.nn.conv2d(x, W, strides = [1, 1, 1, 1], padding = "SAME")

def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize = [1, 2, 2, 1], strides = [1, 2, 2, 1],
padding = "SAME")

W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])

x_image = tf.reshape(x, [-1, 28, 28, 1])

#
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1))
h_pool1 = max_pool_2x2(h_conv1)

#
W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])

h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2))
h_pool2 = max_pool_2x2(h_conv2)

W_fc1 = weight_variable([7764, 1024])
b_fc1 = bias_variable([1024])

h_pool2_flat = tf.reshape(h_pool2, [-1, 7764])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)

#
keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

#
W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])

y_conv = tf.nn.softmax(tf.matmul(h_fc1_drop,W_fc2) +b_fc2)

#
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y_conv), reduction_indices = [1]))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)

correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

init = tf.initialize_all_variables()

saver = tf.train.Saver()

load_model(sess, saver, checkpoint_dir)

for i in range(1):
batch = mnist.train.next_batch(50)
if i%10 == 0:
train_accuracy = accuracy.eval(feed_dict = {x : batch[0] , y_ : batch[1], keep_prob : 1.0})
print("step %d, training accuracy %g"%(i, train_accuracy))

train_step.run(feed_dict = {x : batch[0], y_ : batch[1], keep_prob : 0.5})
print("test accuracy %g"%accuracy.eval(feed_dict={
x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))

tf.scalar_summary("accuracy", accuracy)

saver.save(sess,checkpoint_dir+'model.ckpt')

---

When I restore the checkpoint:

saver.restore(sess, ckpt.model_checkpoint_path)

TensorFlow throws this error:

Traceback (most recent call last):
.
.
.
NotFoundError: Tensor name "global_step_7" not found in checkpoint files ./ckpt_dir/model.ckpt-0
[[Node: save_18/restore_slice_438 = RestoreSlicedt=DT_INT32, preferred_shard=-1, _device="/job:localhost/replica:0/task:0/cpu:0"]]
Caused by op 'save_18/restore_slice_438', defined at:
File "/home/m/anaconda3/lib/python3.5/site-packages/spyderlib/widgets/externalshell/start_ipython_kernel.py", line 205, in
ipythonkernel.start()
.
.
.
File "/home/m/anaconda3/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 1224, in __init
raise TypeError("Control input must be an Operation, "

---

EDIT:

I use anaconda. The first time I run this code in spyder or ipython with "run filename.py", it saves the model in the checkpoint, but when I run this code again, it throws the error.

But when I close spyder or ipython, open it again and run the code it restores the checkpoint correctly.

Also when I run in terminal "python filename.py" it always runs and doesn't throw any error.

Answer 1

You need to reset the default graph at the beginning of your call when you run again the file.

---

If you don't reset the default graph, and run two times the line:

x = tf.Variable(1, name='x')
print x.name

You will see the first time that x has name "x:0" and the second time its name is "x_1:0". This is what confuses tf.train.Saver:

• it first saves the value of x using name "x:0"
• then in the next run you try to load the saved value of x, but now the name of the variable is "x_1:0", so the saver tries to load a saved value under the name "x_1:0" but cannot find it, and returns an error.

---

However, you can reset the default graph at the beginning using tf.reset_default_graph(). This will create an empty graph and use it as default graph.
Here the name of x can be the same in those two graphs:

# First run
tf.reset_default_graph()
x = tf.Variable(1, name='x')
print x.name  # prints 'x:0'

# Next run
tf.reset_default_graph()
x = tf.Variable(1, name='x')
print x.name  # prints 'x:0'

The two Variables can now have the same name because they are no longer in the same graph.

---

Another way of doing it is to create a graph at the beginning and use it as default graph:

graph = tf.Graph()
with graph.as_default():
    x = tf.Variable(1, name='x')