ModuleNotFoundError: No module named 'keras._tf_keras

[michaelbcerny]

I enjoyed reading your paper in Political Analysis, and I'm now trying to implement the 'rMIDAS' package in R. I'm encountering a ModuleNotFoundError: No module named 'keras._tf_keras' error when attempting to use TensorFlow and TensorFlow Addons via the reticulate package in R. This issue persists despite setting up a clean Python environment and ensuring that compatible versions of TensorFlow and TensorFlow Addons are installed.

My specifications:

• Platform: Posit Cloud
• R version: 4.3.3
• Python version: 3.11.4
• TensorFlow version: 2.16.1 (also attempted with 2.15.0)
• TensorFlow Addons version: 0.20.0

To recreate the error:

1. Create a new virtual environment using reticulate::virtualenv_create(envname = "rmidas_env", python_version = "3.8")
2. Install TensorFlow and TensorFlow Addons:
   • reticulate::py_install("tensorflow==2.15.0", envname = "rmidas_env")
   • reticulate::py_install("tensorflow-addons==0.20.0", envname = "rmidas_env")
3. Attempt to import TensorFlow and TensorFlow Addons in R:
   • reticulate::py_run_string("import tensorflow")
   • reticulate::py_run_string("import tensorflow_addons")

The issue seems related to how TensorFlow Addons interacts with the TensorFlow/Keras integration (adjustments to the versions and reinstallation of packages did not resolve the issue).

Do you have any suggestions?

Thank you for looking into this!