Tensorflow program results in type conversion error

Question

Tring to master Tensorflow, following documentation of TensorFlow.

Below program results in 'Incompatible type conversion error'

import tensorflow as tf

W = tf.Variable([.3], tf.float32)
b = tf.Variable([-3], tf.float32)
x = tf.placeholder(tf.float32)
linear_model = 1.0
linear_model = W * x + b
#tf.to_float(linear_model, name='ToFloat')

# Global initialization is must
init = tf.global_variables_initializer()
sess.run(init)

print(sess.run(linear_model, {x:[1,2,3,4]}))

Above program results in this error

File "v-prog3-variables.py", line 7, in linear_model = W * x + b .. .. .. ValueError: Incompatible type conversion requested to type 'float32' for variable of type 'int32_ref'

I tried to solve the problem by defining the 'linear_model' variable as float (linear_model = 1.0) or tf.to_float(linear_model = W * x + b)

but nothing works

Im a TensorFlow newbie, please help me out. Thanks in advance.

Answer 1

I was able to get it to run by recasting it to float32. Have you read the source code of the library? Yea... I don't ask questions lol

import tensorflow as tf

W = tf.Variable([.3], tf.float32)
b = tf.Variable([-3], tf.float32)
x = tf.placeholder(tf.float32)

linear_model = W * x + tf.cast(b, tf.float32)

init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)

print(sess.run(linear_model, {x:[1,2,3,4]}))

Answer 2

You need to use a named argument for the type, e.g. tf.Variable([.3], dtype=tf.float32)

The signature of tf.Variable is __init__(self, initial_value=None, trainable=True,... leading to common errors.

The __init__ method will then infer type from your inputs:

• [.3] will give a tf.float32, and
• [-3] will give a tf.int32 leading to the error you got when you multiply them.

If you want to stick with the tf.float32 type, you could also use [-3.] as initial value for b.