Tensorflow python. ValueError: Non-scalar tensor cannot be converted to boolean

Question

I have been trying out these codes related to tensorflow 1.12.2 install together with visual studio 15.9.6. The python version is 3.6.6.

The problem lies in the conditional statement in the log_huber function. Any advise on how to solve this is greatly appreciated. The code is appended below:

import tensorflow as tf
import numpy as np

def log_huber(x, m):
  if tf.abs(x) <= tf.abs(m):
    return x**2
  else:
    return m**2 * (1 - 2 * tf.log(m) + tf.log(x**2))

x = np.arange(10,dtype=np.float32)
m = np.arange(10,20,dtype=np.float32)

_x = tf.data.Dataset.from_tensor_slices(x).shuffle(10).repeat().batch(1)
iter_x = _x.make_one_shot_iterator()
_x_init_ops = iter_x.make_initializer(_x)
next_x = iter_x.get_next()

_m = tf.data.Dataset.from_tensor_slices(m).shuffle(10).repeat().batch(1)
iter_m = _m.make_one_shot_iterator()
_m_init_ops = iter_m.make_initializer(_x)
next_m = iter_m.get_next()

y = tf.contrib.eager.py_func(func=log_huber, inp=[next_x,next_m], Tout=tf.float32)
with tf.Session() as sess:
    sess.run([_x_init_ops,_m_init_ops])

    terminate = 1

    while terminate!="0":
        print(sess.run(y))
        terminate = input("0 for end, 1 to continue")

The error message is as follow:

...\testTensorboard\testTensorboard\dataset.py", line 5, in log_huber
    if tf.abs(x) <= tf.abs(m):

 ...\conda\conda\envs\rdkit-env\lib\site-packages\tensorflow\python\framework\ops.py", line 914, in __bool__
    "Non-scalar tensor %s cannot be converted to boolean." % repr(self))

ValueError: Non-scalar tensor <tf.Tensor: id=58, shape=(1,), dtype=bool, numpy=array([False])> cannot be converted to boolean.

.eval() function can also help. This evaluates the expression and returns a value — shantanu pathak
– shantanu pathak, Commented Jun 20, 2019 at 17:04

Mohan Radhakrishnan · Accepted Answer · 2019-03-04 14:19:17Z

3

If you use tf.squeeze like this your dimensions are removed.

def log_huber(x, m):
  print (tf.abs(x))
  if tf.squeeze(tf.abs(x)) <= tf.squeeze(tf.abs(m)):
    return x**2
  else:
    return m**2 * (1 - 2 * tf.log(m) + tf.log(x**2))

It removes dimensions of size 1 from the shape of this tensor

tf.Tensor([2.], shape=(1,), dtype=float32)

edited Mar 4, 2019 at 14:19

answered Mar 4, 2019 at 9:46

Mohan Radhakrishnan

3,2385 gold badges35 silver badges46 bronze badges

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

Collectives™ on Stack Overflow

Tensorflow python. ValueError: Non-scalar tensor cannot be converted to boolean

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related