3 ms·
Understanding the bias-variance tradeoff
- stygiansonic 11y agoOf note, figure 3 is also known as a Voronoi diagram: https://en.wikipedia.org/wiki/Voronoi_diagram https://en.wikipedia.org/wiki/Voronoi_diagram
- jdonaldson 11y agoWe use so many phrases to represent the same concepts : https://en.wikipedia.org/wiki/Accuracy_and_precision https://en.wikipedia.org/wiki/Accuracy_and_precision
- argonaut 11y agoUnfortunately, as you can see on the wiki article, accuracy and precision have a different meaning for binary classifiers.
- jdonaldson 11y agoMeanwhile, variance has a different meaning in nearly every other statistical context. Confusion abounds. It gets worse with acronyms. Sigh... I'm just going on a rant, really.
- adenadel 11y agoWhat do you mean? The use in this context is the definition of variance: Var(X) = E[(X - E[X])^2]
- choxi 11y agoI notice this sometimes too. I think another example is "burnout" and "learned helplessness" [1]. Whenever I see another article on HN trying to understand what burnout is, I feel like we're rediscovering something psychologists discovered in the 60's. [1] https://en.wikipedia.org/wiki/Learned_helplessness https://en.wikipedia.org/wiki/Learned_helplessness
- johannesg 11y agoFunny, these terms are used in my field: https://en.wikipedia.org/wiki/Validity_(statistics) https://en.wikipedia.org/wiki/Validity_(statistics) and https://en.wikipedia.org/wiki/Reliability_(psychometrics) https://en.wikipedia.org/wiki/Reliability_(psychometrics). Terms aside, this is a very good article. I've experienced the trade-off between reliability and validity in devising measurements myself, but had not realized it is unavoidable.