I’m doing the C1W3 assignment which requires Tensorflow 2.2 and ran into a number of version compatibility issues, with errors and runaway processes. I burned a few hours and hope this might help someone in the future.
I’m running on my local PC with Windows 10 and Anaconda. My existing Anaconda environments are all using Python 3.9 and Tensorflow-GPU 2.6.
Cloning one of the working environments and migrating to the older versions failed.
Creating a new Anaconda Environment with Python 3.9 and then installing Tensorflow 2.1 or 2.3 appeared to work, but the kernel would lock up the CPU when I tried to build even a one node model.
To get a working Tensorflow 2.2 environment I had to :
- Create a new Anaconda environment with Python 3.7
- Use Anaconda to install Tensorflow 2.1. ( I didn’t bother with the GPU )
- Use pip to upgrade Tensorflow to 2.2. ( Anaconda can’t install 2.2 )
- Use pip to install tensorflowjs
- Use Anaconda to uninstall Pillow and then install Image
Not hugely complex, but iterations are slow and it took quite a while to find a working combination.