Fix CI failures

[schallerdavid]

Description

This PR aims to fix CI failures involving RDKit installation on MacOS and T018 with failing API request of DogSiteScorer.

Todos

• fix RDKit installation on MacOS (pin fontconfig)
• fix T018 with failing API requests of DogSiteScorer
• fix tensorflow AttributeError only seen for MacOS 44467 (pin h5py)
• seeing dead nb kernel for MacOS

Status

• Ready to go

[schallerdavid]

conda-forge/fontconfig-feedstock#54

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[schallerdavid]

Python 3.7 succeeded for f7869fd. However, the other versions failed for various reasons including MacOS with an error related to tensorflow. I pushed a new commit pinning the h5py version, but now GitHub actions got stuck. I gotta take care of a presentation now and will check later.

[schallerdavid]

I added two NBVAL_SKIP statements to the cells in T018 doing requests for the dogsitescorer api based on an uploaded PDB structure. This seems to fail currently, maybe that's temporary, maybe for a longer time period, but it's likely a problem from protein.plus.

[schallerdavid]

The nb kernel dies on MacOS for some reason, letting it rerun to see if it only happens occasionally.

[schallerdavid]

Btw, I also played with multiprocessing of nbval for KinoML. I often got more stable runs, when only using a single process.

[schallerdavid]

So all tests pass 🎉, I will merge now to get the CI working in master, but it is very concerning that different CI runs fail for different reasons. I will open an issue tomorrow, so we dont forget to investigate this.