Applying software engineering principles to Machine Learning projects
Login to save
07:00 PM – 08:10 PM IST
Rohit asks: For workflows that use jupyter notebooks (or any other type of notebooks), how do you write unit tests for it? And a more general question - how do you write unit tests for data science?
Dmitry: Jupyter - is for exploration mostly. It is ok to have testing code right in the notebook in the exploration phase. If you reached the point of writing unit-test for ML code it can be a good sign to switch from exploration mode and Jupiter to writing code in text files.
Ivan: Yep, agreed. But if you indeed want to test notebooks, it can be done probably with the nbconvert (https://nbconvert.readthedocs.io/en/latest/execute_api.html#module-nbconvert.preprocessors) tool or papermill to parametrize and run them as a script.
Login to leave a comment