[PYTHON] A story about an error when loading a TensorFlow model created with Google Colab locally

environment

Google Colab environment

Python3.6.9 TensorFlow2.3.0

Local environment

Python3.6.7 TensorFlow2.1.0

what's happened?

Download the model created by Google Colab and When I tried to run it in the local environment, I got angry when TesorFlow was imported.

ʻImportError: DLL load failed: The specified module cannot be found `

Microsoft Visual C ++ Redistributable required from TensorFlow 2.1

After investigating, it seems that Microsoft Visual C ++ Redistributable is required from TensorFlow 2.1. Install from the following URL. https://support.microsoft.com/en-us/help/2977003/the-latest-supported-visual-c-downloads

This is the solution! However, I encounter an error again.

I was able to import, but when I loaded the model, I got an error saying KeyError:'sample_weight_mode.

I tried to load the model with the following code ...

from tensorflow.keras.models import load_model
new_model = load_model(r'***.h5')

A message like ↓ is displayed and cannot be read. KeyError: 'sample_weight_mode'

Once again, thanks to Google Sensei, There was a person with the exact same phenomenon in an overseas forum. https://github.com/keras-team/keras/issues/14040

According to this article, it seems that an error will occur when trying to load a model created with TensorFlow 2.3 with a lower version.

When I raised TensorFlow in my local environment to 2.3, it loaded successfully. (The command is below)

pip install tensorflow==2.3.0