CODEWITHSUNDEEP
pythontensorflow
rintiunse
rintiunse

Reputation: 23

TensorFlow the variable cant be initialized

X=tf.placeholder(tf.float32,[None,32,32,3])
y=tf.placeholder(tf.int64,[None])
is_training=tf.placeholder(tf.bool)

def simple_model(X,y):

    Wconv1=tf.get_variable("Wconv1",shape=[7,7,3,32],use_resource=True)
    bconv1=tf.get_variable('bconv1',shape=[32])
    W1=tf.get_variable('W1',shape=[5408,10])
    b1=tf.get_variable('b1',shape=[10])

    a1=tf.nn.conv2d(X,Wconv1,[1,2,2,1],'VALID')+bconv1
    h1=tf.nn.relu(a1)

    h1_flat=tf.reshape(h1,[-1,5408])
    y_out=tf.matmul(h1_flat,W1)+b1
    return y_out

init=tf.global_variables_initializer()

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

    sess.run(simple_model(X,y),feed_dict={X:X_train,y:y_train})

error is

PreconditionError Attempting to use uninitialized variable Wconv1

I dont know what wrong with code?

Upvotes: 2

Views: 33

Answers (1)

javidcf
javidcf

Reputation: 59681

tf.global_variables_initializer makes an initialization op for all the global variables created up to that point. This means that if you create other variables later, they will not be initialized by the operation. This is because variable initializers just hold a list of the variables they have to initialize, and this does not change as you add more variables (in fact, tf.global_variables_initializer() is just a shortcut for tf.variables_initializer(tf.global_variables()) or tf.variables_initializer(tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES))). In your case, the variables are being created in the second call to sess.run, after you create init earlier. You need to create the initialization operation after you have created your model with the variables:

X=tf.placeholder(tf.float32,[None,32,32,3])
y=tf.placeholder(tf.int64,[None])
is_training=tf.placeholder(tf.bool)

def simple_model(X,y):

    Wconv1=tf.get_variable("Wconv1",shape=[7,7,3,32],use_resource=True)
    bconv1=tf.get_variable('bconv1',shape=[32])
    W1=tf.get_variable('W1',shape=[5408,10])
    b1=tf.get_variable('b1',shape=[10])

    a1=tf.nn.conv2d(X,Wconv1,[1,2,2,1],'VALID')+bconv1
    h1=tf.nn.relu(a1)

    h1_flat=tf.reshape(h1,[-1,5408])
    y_out=tf.matmul(h1_flat,W1)+b1
    return y_out

my_model = simple_model(X,y)
init=tf.global_variables_initializer()

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

    sess.run(my_model, feed_dict={X:X_train,y:y_train})

Upvotes: 1

Related Questions