Run changed notebooks in CI

[jhamon]

Problem

We need to ensure that notebooks which are heavily featured in our docs are actually working

Solution

Implement a workflow to convert a list of specified notebooks into scripts that can be run in CI. For each notebook (which is a json document with an *.ipynb file extension):

• Create a temporary directory to isolate from other parts of the repo. Copy the single notebook over to that tmpdir.
• Use the nbformat packge to iterate through notebook cells
• If a cell is showing pip install steps, include those in a run.sh bash setup script to activate and configure a venv with required dependencies
• Collect code from other code cells into a python script called notebook.py
• From the workflow, run the run.sh script to install needed dependencies and execute the notebook.py script
• Execution without errors considered a success

Since some notebooks are quite slow/lengthy (for example, if computing embeddings across a large sample dataset) and there are a lot of them (over 200 notebooks in this repo), we won't run every notebook on every push because that would be slow/expensive and deter people from making small improvements. Instead, we will only run those notebooks which have changed relative to the master branch.

Some notebooks may fail until minor adjustments are made to them to account for implicit dependencies on things installed in the colab environment that are not available in our CI environment. For example, we may need to add a missing a statement like !pip install pandas instead of just using it directly.

Type of Change

• Infrastructure change (CI configs, etc)

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB