How can I convert a tensor into a ndarray in TensorFlow?

Question

My goal is to convert a tensor into a ndarray without 'run' or 'eval'. I wanted to perform the same operation as the example.

A = tf.constant(5)
B = tf.constant([[A, 1], [0,0]])

However, ndarray can be inside tf.constant but tensor cannot. Therefore, I tried to perform the operation using the following example, but tf.make_ndarray does not work.

A = tf.constant(5)
C = tf.make_ndarray(A)
B = tf.constant([[C, 1], [0,0]])

https://github.com/tensorflow/tensorflow/issues/28840#issuecomment-509551333

As mentioned in the github link above, tf.make_ndarray does not work. To be precise, an error occurs because tensorflow requires a 'tensor_shape' that does not exist, instead of a 'shape' that exists.

How can I run the code in this situation?

tf.make_ndarray expects a TensorProto (e.g. read from a serialized graph or made with tf.make_tensor_proto), not a tf.Tensor. — jdehesa

jdehesa jdehesa · Accepted Answer · 2019-08-13T09:18:26

tf.make_ndarray is used to convert TensorProto values into NumPy arrays. These values are generally the constants used in a graph. For example, when you use tf.constant, you create a Const operation with an attribute value holding the constant value that the operation will produce. That attribute is stored as a TensorProto. Hence, you can "extract" the value of a Const operation as a NumPy array like this:

import tensorflow as tf

A = tf.constant(5)
C = tf.make_ndarray(A.op.get_attr('value'))
print(C, type(C))
# 5 <class 'numpy.ndarray'>

In general, though, you cannot convert arbitrary tensors into NumPy arrays, as their values will depend on the values of the variables and the fed inputs within a particular session.

How can I convert a tensor into a ndarray in TensorFlow?

1 Answers