Author: Christoph Molnar
This book is the result of an experiment in university teaching. Each semester, students of the Statistics Master can choose from a selection of seminar topics. Usually, every student in the seminar chooses a scientific paper, gives a talk about the paper and summarizes it in the form of a seminar paper. The supervisors help the students, they listen to the talks, read the seminar papers, grade the work and then … hide the seminar papers away in (digital) drawers. This seemed wasteful to us, given the huge amount of effort the students usually invest in seminars. An idea was born: Why not create a book with a website as the outcome of the seminar? Something that will last at least a few years after the end of the semester. In the summer term 2019, some Statistics Master students signed up for our seminar entitled “Limitations of Interpretable Machine Learning”. When they came to the kick-off meeting, they had no idea that they would write a book by the end of the semester.
We were bound by the examination rules for conducting the seminar, but otherwise we could deviate from the traditional format. We deviated in several ways:
- Each student project is part of a book, and not an isolated seminar paper.
- We gave challenges to the students, instead of papers. The challenge was to investigate a specific limitation of interpretable machine learning methods.
- We designed the work to live beyond the seminar.
- We emphasized collaboration. Students wrote some chapters in teams and reviewed each others texts.
Looking back, the seminar was a lot of fun and – from our perspective – successful. Especially considering that it was an experiment. Everyone was highly motivated and we got great feedback from the students that they liked the format. For the students it was a more work than a traditional seminar. But in the end, our hope is that their effort will pay off for them as well, not only because of their increased visibility. It was also more work for us supervisors. But the extra effort was worth it, since limitations of interpretability are relevant for our research. For me the seminar was an inspiration. The students had new ideas and new perspectives to approach the limitations of interpretable machine learning.
The book chapters are written in the Markdown language. The simulations, data examples and visualizations were created with R (R Core Team 2018). To combine R-code and Markdown, we used rmarkdown. The book was compiled with the bookdown package. We collaborated using git and github. For details, head over to the book’s repository.
R Core Team. 2018. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/.