tflearn import issue about tflearn

[rela-v]

Hi! Thanks for your question.

tl;dr: everywhere you say "is_sequence" in your code, replace it with "is_sequence_or_composite".

Your problem

• If you try to do the following:
  from tensorflow.python.util.nest import is_sequence
• The following error results:
  ImportError: cannot import name 'is_sequence' from 'tensorflow.python.util.nest'
(/Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/site-packages/tensorflow/python/util/nest.py)

Explanation of the Problem

If this is indeed the problem, then I encountered the same issue several days ago and was able to resolve it. The reason you are experiencing the error is that, as the error message says, you
cannot import name 'is_sequence' from 'tensorflow.python.util.nest'
which means that there was some issue in the import process. Indeed, cursory research reveals that there is no longer any name "is_sequence" in the nest.py module (per tensorflow/tensorflow@f564400).

Solution

If you visit the provided link above, there is a new function which will serve your purpose well named "is_sequence_or_composite()", which according to the source code does the same exact thing as "is_nested_or_composite()" (use whichever one floats your boat!).
Thus, replacing the "is_sequence" function with the "is_sequence_or_composite" function will resolve the error.
For more information on what a "nested structure" is, which is what this function actually checks for, see the following link:
https://www.tensorflow.org/api_docs/python/tf/nest

Thank you so much again for asking. I hope this works for you, and feel free to ask for any help you may need along the way!

-Rela

[MihaMarkic]

@rela-v It happens if one doesn't use is_sequence explictly, simple import tflearn is enough to trigger the error. So, it is not possible to correct the error which is most probably somewhere in tflearn code.

[MihaMarkic]

And as expected, is_sequence is imported in tflearn/layers/recurrent.py file.

[MihaMarkic]

Created a PR

[awsumbill]

In the meantime you can pip install tensorflow<2.13 to get around this untill it's fixed