Trust but verify - testing optimization models in real life

Speaker: Richard Oberdieck, Banking Circle

"Unless you magically find yourself endowed with limitless programming support, sooner or later (most likely sooner) you are likely to need to do some coding." (Paul Rubin)

Between a beautiful model on paper and real world impact lies the big gap of implementation. To bridge this gap confidently, automated tests are a most important pillars on top of which your code rests. However, many optimization professionals are not familiar with the ins and outs of automated testing. What makes this even more difficult is that writing tests for optimization models specifically is not trivial. Between licensing issues when one uses commercial solvers to run times being many hours, it is not easy to find a good approach on how to build these solid pillars.In this talk, I will share what I have learned about testing optimization models over the past 8 years. We will discuss the anatomy of a good test, what types of testing exists and how property-based testing combined with ChatGPT and good data structures can provide a robust testing setup for code involving optimization models.

About the speaker:

Since getting his PhD from Imperial College in 2017, Richard Oberdieck has spent his career at the intersection between code and optimization. First at Ørsted as a software developer for a cable design tool, then at Gurobi as a technical account manager helping countless customers getting their optimization projects off the ground. Currently, he works for a bank called Banking Circle as a data scientist, where he works on a tool which, if it breaks, stops all payments in the bank, leading to an even higher focus on testing than before.