cannot import name 'deserialize_keras_object' from partially initialized module 'keras.saving.legacy.serialization'most likely due to a circular impor

Question

I was trying to run simple Machine Learning algorithm on Visua Studio Code. I downloaded Tensorflow 2 and installed it. Code I was trying to run was:

import tensorflow as tf
import numpy as np
from tensorflow import keras
model = tf.keras.Sequential([keras.layers.Dense(units=1, input_shape=[1])])

As I understood this shows how machine learning model works in general.

Then on line 4 it says

Exception has occurred: ImportError"
"cannot import name 'deserialize_keras_object' from partially initialized module 'keras.saving.legacy.serialization' (most likely due to a circular import) (C:\Users\Erdemdavaa\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.9_qbz5n2kfra8p0\LocalCache\local-packages\Python39\site-packages\keras\saving\legacy\serialization.py)
  File "D:\study\semester 6\Project laborotory(team)\ML_experiment.py", line 4, in <module>
    model = tf.keras.Sequential([keras.layers.Dense(units=1, input_shape=[1])])
ImportError: cannot import name 'deserialize_keras_object' from partially initialized module 'keras.saving.legacy.serialization' (most likely due to a circular import) (C:\Users\Erdemdavaa\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.9_qbz5n2kfra8p0\LocalCache\local-packages\Python39\site-packages\keras\saving\legacy\serialization.py)

What is this and how do I fix this?

Answer 1

I recently have a similar issue. It is it probably a compatibility issue between keras and tensorflow. A temporary solution is to use the tensorflow.python module, (although it is experimental, it should suffice a small machine learning project).

import tensorflow as tf
from tensorflow.python.keras import layers
from tensorflow.python.keras.models import Sequential

Sequential is imported so you can directly use it.

Answer 2

Had the same issue and at first I was unable to resolve it. What eventually worked for me was the following (Note this appears to be a predominant problem on machines with the M1/M2 chip):

I created a virtual environment folder using the Terminal where I will store all environments. Later you can pass it to VS Code:

$ pip install virtualenv
$ cd ~
$ mkdir .virtualenvs
$ cd .virtualenvs

Inside the newly created directory, you can create your environment and activate it:

$ virtualenv VENV_NAME
$ source VENV_NAME/bin/activate

In the VS Code preferences you can search for 'venv' and set the 'Python: Venv Path' option to ~/.virtualenvs (or whatever you decided to call your environment folder). Then I was able to select the newly created environment as a kernel. You might need to add it by searching for 'Python: Select Interpreter' after pressing (Cmd + Shift + P) - you only need this step if you are planning to use the environment in VS Code.

After these steps I simply installed tensorflow according to the apple developer page (https://developer.apple.com/metal/tensorflow-plugin/):

$ python -m pip install tensorflow

And tensorflow-metal which allows for training on Mac GPUs:

$ python -m pip install tensorflow-metal

This allowed me to run tensorflow without any issues. Hope this might resolve the issue for some.