"ValueError: setting an array element with a sequence." TensorFlow

Question

I tried to explore a simple demo，but got this error. How could I modify my code?

import tensorflow as tf

sess = tf.Session()

x_ = tf.Variable([[-9,6,-2,3], [-4,3,-1,10]], dtype=tf.float32)
x = tf.placeholder(tf.float32, shape=[4,2])
y = tf.nn.relu(x)

sess.run(tf.global_variables_initializer())
print(sess.run(x_))
print(sess.run(y,feed_dict={x:x_}))

The output I get is:

[[ -9.   6.  -2.   3.]
 [ -4.   3.  -1.  10.]]

Traceback (most recent call last):
  File "C:\Users\jy\Documents\NetSarang\Xftp\Temporary\test.py", line 19, in <module>
    print(sess.run(y,feed_dict={x:x_}))
  File "C:\Program Files\Anaconda3\lib\site-packages\tensorflow\python\client\session.py", line 767, in run
    run_metadata_ptr)
  File "C:\Program Files\Anaconda3\lib\site-packages\tensorflow\python\client\session.py", line 938, in _run
    np_val = np.asarray(subfeed_val, dtype=subfeed_dtype)
  File "C:\Program Files\Anaconda3\lib\site-packages\numpy\core\numeric.py", line 482, in asarray
    return array(a, dtype, copy=False, order=order)
ValueError: setting an array element with a sequence.

Michael0x2a · Accepted Answer · 2017-07-09 04:22:38Z

1

First, the dimensionality of your x_ variable wrong: currently, it's of shape [2, 4], but you're attempting to use it in a slot that's expecting data of shape [4, 2].

Second, tf.Variable is meant to represent literally a variable (in the mathematical sense) within your neural net model that'll be tuned as you train your model -- it's a mechanism for maintaining state.

To provide actual input to train your model, you can simply pass in a regular Python array (or numpy array) instead.

Here's a fixed version of your code that appears to do what you want:

import tensorflow as tf

sess = tf.Session()

x = tf.placeholder(tf.float32, shape=[4,2])
y = tf.nn.relu(x)
sess.run(tf.global_variables_initializer())

x_ = [[-9, -4], [6, 3], [-2, -1], [3, 10]]
print(sess.run(y, feed_dict={x:x_}))

If you really did want a node within your neural net to start off initialized with those values, I'd get rid of the placeholder and use x_directly:

import tensorflow as tf

sess = tf.Session()

x = tf.Variable([[-9, -4], [6, 3], [-2, -1], [3, 10]], dtype=tf.float32)
y = tf.nn.relu(x)
sess.run(tf.global_variables_initializer())

print(sess.run(y))

This is probably not what you meant to do, though -- it's sort of unusual to have a model that doesn't accept any input.

answered Jul 9, 2017 at 4:22

Michael0x2a

65.6k32 gold badges189 silver badges241 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

jinyangzhang Over a year ago

Actually，I want to tempt a new active function after a sigmoid function，which changes the input X when X>6 , output y=1；X<-6,y=0；otherwise,don't change X's elements , how I could do? I now the sigmoid don't effective like ReLU，I just want to try something.

jinyangzhang Over a year ago

(Sorry for the virgin question on stackoverflow)Actually，I want to tempt a new active function after a sigmoid function，which changes the input X when X>6 , output y=1；X<-6,y=0；otherwise,don't change X's elements , how I could do? I know the sigmoid don't effective like ReLU，I just want to try something.I'm test it on mnist dataset.

Collectives™ on Stack Overflow

"ValueError: setting an array element with a sequence." TensorFlow

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related