train_autoencoder.ipynb error I got

[HalitTalha]

Hi there,
Sorry in advance if this sounds silly and has already been addressed, I'm trying to get used to colab and dependency conflicts etc. When I try to install, on the first step in the colab demo, I'm getting a "metadata generation failed" error.
I'm not changing runtime or any other configuration. By any chance, could you point me in the right direction?

It's the same when I try it locally. I figure this might be related to the runtime and pip versions, which are the latest in my case.

[markhanslip]

I've found that if you run the following cell and restart the runtime once it completes - before running an updated install command - it at least gets you some of the way.

First run:

!pip uninstall -y keras tensorflow tensorflow-probability absl-py astunparse flatbuffers gast google-pasta grpcio h5py keras keras-preprocessing libclang numpy opt-einsum protobuf setuptools six tensorboard tensorflow-io-gcs-filesystem termcolor tf-estimator-nightly typing-extensions wrapt
!pip install --disable-pip-version-check --no-cache-dir tensorflow==2.11.0
!pip install tensorflow-probability==0.15.0
!pip install keras==2.11.0

Then click on the 'restart runtime' button that appears in the output from the above cell. Then run:

# !pip install -qU ddsp[data_preparation]==1.6.3

!sudo apt-get install libportaudio2 &> /dev/null
!pip install -U ddsp[data_preparation] &> /dev/null

# Initialize global path for using google drive.
DRIVE_DIR = ''

This way at least the base packages are installed. However, the train_autoencoder notebook then fails further down when trying to compute the dataset statistics, which is an issue several others have commented on and no one seems to have a good fix for.

The above fix works for the train VST notebook (https://github.com/magenta/ddsp-vst) and that notebook does allow you to train a model and export a plugin.

In case someone from Magenta is checking this, it'd be nice if the notebooks actually worked!