The Apothecary Diaries is my favourite anime. While choosing a new name for the conference section of this blog, I picked “Conference Diaries” because it reminded me of it. And then I realised something: Maomao, its heroine, would make an excellent data scientist.
(Mild spoiler of the first episodes ahead.)
When the babies of the imperial palace fall ill, everyone talks about a curse. Maomao does not. She looks at what the sick children have in common, and finds it in the white face powder their mothers use, which contains lead. No curse, just a pattern nobody had looked for.
I think about her every time a result looks mysterious. When a model works beautifully in one hospital and badly in another, it is rarely a curse. Maybe the second hospital uses a different MRI scanner, or sees older patients. The first question is always Maomao’s: what do these cases have in common?
She also distrusts explanations that are too convenient, and checks everything herself. That is exactly the attitude I want in front of a model that looks too good to be true.
Here, however, the analogy has to stop: Maomao tests poisons on herself. Please test your models on independent patients instead.
The pill: be a little like Maomao. Curious, sceptical, and always looking for the pattern before blaming the curse. Just skip the poison tasting.
Leave a Reply