tensorflow from_tensor_slices gives "args_0:0"

Question

I'm trying to create a tensorflow dataset using from_tensor_slices but the string I'm passing is getting messed up within the map function:

def test(x):
    print(x)
    return x

filenames=['a','b','c']


dataset = tf.data.Dataset.from_tensor_slices(filenames)
dd= dataset.map(test) 

The function prints:

Tensor("args_0:0", shape=(), dtype=string)

When I try:

print(next(dd.as_numpy_iterator()))

It gives "a" as expected. But I want to access the string inside the function so that it can open filenames. What's going on? I found a similar answer here: "args_0:0" when printing Tensor But it wasn't enlightening and was not a minimal example. I also want to know the solution for from_tensor_slices I checked tf.executing_eagerly() and it gives True. My tensorflow version is 2.15.1. I'd prefer not to upgrade.

Answer 1

I found a similar answer here: Tensorflow Dataset open several files and keep them separated

Apparently only tensorflow functions can be applied to tensorflow arguments within the map function. So my test function would work if I replaced print(x) with tf.print(x).