ModuleNotFoundError: No module named 'tensorflow.python.trackable'

Question

The statement is:

from tensorflow.python.trackable import base as trackable

But the module not found error is coming up.

##### File "C:\Users\mahit\anaconda3\lib\site-packages\keras\dtensor\lazy_variable.py", line 26, in <module>
   from tensorflow.python.trackable import base as trackable

ModuleNotFoundError: No module named 'tensorflow.python.trackable'

The piece of code is inside keras library. I dont understand why this module not found error is coming up. Becase this is a code that came with the package. Also not sure what trackable does. Please help me resolve this.

Answer 1

I have the same issue and have been check the currenct dir of tensorflow, the trackable folder does not exist.

So here is an easy way.

1. Git clone from the url: https://github.com/tensorflow/tensorflow
2. Find the tensorflow -> python -> trackable
3. Copy the trackable folder to your tensorflow install dir (For example: C:\Users\xxxx\anaconda3\Lib\site-packages\tensorflow\python ).

Answer 2

I got rid of this error: uninstall tensorflow and install tf-nightly. Nevertheless, I got another error: Can't find libdevice directory ${CUDA_DIR}/nvvm/libdevice

In python code, with Linux OS, that error could be solved by using command: os.environ['XLA_FLAGS'] = '--xla_gpu_cuda_data_dir=/usr/lib/cuda/'

Answer 3

I had the same problem for the past one month and I was not able to resolve it. Then finally I just uninstalled tensorflow and installed tf-nightly with immediately resolved the issue.

The issue occurs as tensorflow recently moved the keras out of the default package and now is only available through the tf-nightly install.

Hope this resolves the issue

Answer 4

Uninstall tensorflow using pip uninstall tensorflow and install pip install tf-nightly

This fix worked for me perfectly well.

Answer 5

I just got this error in CI builds with a requirements.txt that had

keras
tensorflow==2.9.1

and I assume the error was caused by pip install -r requirements.txt taking the recently released Keras 2.10 version, which probably assumes it's running next to TensorFlow 2.10, and that the minor mismatch led to this import error.

Answer 6

The solution for me was to install keras in version 2.9.0

pip install keras==2.9.0