How to create a tensor from another tensor like tf.constant and number?

Question

I want to use the value in a tensor to create another tensor, but I got the following error:

>>> a = tf.constant(3)
>>> a
Out[51]: <tf.Tensor: shape=(), dtype=int32, numpy=3>
>>> tf.constant([a, 2])
Traceback (most recent call last):
  File "/Users/belter/miniconda3/envs/deside_obj/lib/python3.8/site-packages/IPython/core/interactiveshell.py", line 3369, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-53-7af9a5175a59>", line 1, in <cell line: 1>
    tf.constant([a, 2])
  File "/Users/belter/miniconda3/envs/deside_obj/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py", line 267, in constant
    return _constant_impl(value, dtype, shape, name, verify_shape=False,
  File "/Users/belter/miniconda3/envs/deside_obj/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py", line 279, in _constant_impl
    return _constant_eager_impl(ctx, value, dtype, shape, verify_shape)
  File "/Users/belter/miniconda3/envs/deside_obj/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py", line 304, in _constant_eager_impl
    t = convert_to_eager_tensor(value, ctx, dtype)
  File "/Users/belter/miniconda3/envs/deside_obj/lib/python3.8/site-packages/tensorflow/python/framework/constant_op.py", line 102, in convert_to_eager_tensor
    return ops.EagerTensor(value, ctx.device_name, dtype)
ValueError: TypeError: Scalar tensor has no `len()`
Traceback (most recent call last):
  File "/Users/belter/miniconda3/envs/deside_obj/lib/python3.8/site-packages/tensorflow/python/framework/ops.py", line 1170, in __len__
    raise TypeError("Scalar tensor has no `len()`")
TypeError: Scalar tensor has no `len()`

How can I use the value in tensor a?

Asked By: Belter

||

Source

Answer 1

Call the .numpy() method to get the tensor value


tf.constant([a.numpy(), 2])

Answered By: zousan

Answer 2

You can use tf.stack.

import tensorflow as tf

@tf.function
def join_tns_num(tensor, num):
    return tf.stack([tensor, tf.constant(num)], axis=0)

Check function:

>>> join_tns_num(tf.constant(3), 2)
<tf.Tensor: shape=(2,), dtype=int32, numpy=array([3, 2], dtype=int32)>

Answered By: I'mahdi

Answer 3

I don’t think tf.stack is the best option for this simple operation, and indeed we want to avoid tf.Tensor.numpy() to retain compatibility with graph mode.

Instead, you can use tf.convert_to_tensor, which, citing the documentation, accepts Tensor objects, numpy arrays, Python lists, and Python scalars, so most things you would ever want to throw at it:

In [1]: import tensorflow as tf
In [2]: a = tf.constant(3)
In [3]: tf.convert_to_tensor([a, 2])
Out[3]: <tf.Tensor: shape=(2,), dtype=int32, numpy=array([3, 2], dtype=int32)>

This also works in graph mode. Adapting the demonstration from another answer:

In [4]: @tf.function
   ...: def join_tns_num(tensor, num):
   ...:     return tf.convert_to_tensor([tensor, tf.constant(num)])
   ...: 
In [5]: join_tns_num(a, 42)
Out[5]: <tf.Tensor: shape=(2,), dtype=int32, numpy=array([ 3, 42], dtype=int32)>

So, if you encounter ValueError: TypeError: Scalar tensor has no `len()`, check whether replacing tf.constant by tf.convert_to_tensor is the answer.

Answered By: fuenfundachtzig

How to create a tensor from another tensor like tf.constant and number?

Question:

Answers: